1. Background and significance

Autism Spectrum Disorder (ASD): Autism is a lifelong neurodevelopmental disorder characterized by difficulties in social communication, language development, and stereotyped repetitive behaviors. According to the latest statistics from the US Centers for Disease Control (CDC), the global incidence of autism has reached about 2.3% (i.e., 1 in 44 children). A large number of children with autism have severe impairments in social interaction, emotional communication, and cognitive development, and about 70% of individuals with autism are unable to live independently throughout their lives. Autism not only brings a heavy psychological and economic burden to the patient's family, but also puts great pressure on social and public resources. It is estimated that annual spending on special education, care and care related to autism will climb to $5.5 trillion by 2060. In view of the high incidence of autism and the resulting burden on families and society, countries have listed overcoming autism as a major public health and scientific and technological strategic need, and there is an urgent need to develop new technologies to improve the quality of life of people with autism.

National Strategic Needs and Scientific Research Frontiers: In China, the early screening, diagnosis and intervention of autism are also facing great challenges. At present, clinical practice mainly relies on empirical scale assessment and behavioral observation, with strong subjective factors, lagging diagnosis, and lack of professional resources in rural and underdeveloped areas, and many children miss the golden period of early intervention. With the advancement of major national projects such as "brain science and brain-like research", the mechanism research and the development of new diagnosis and treatment technologies for developmental disorders, especially autism, have been listed as one of the priority areas. The international trend of autism research shows that it is necessary to integrate multidisciplinary methods to tackle key problems in the whole chain from genetic molecules to neural circuits to behavioral cognition. Specifically, on the one hand, brain imaging and neuromodulation technologies have revealed that patients with autism have abnormal connections in brain regions related to social cognition, such as the dysfunction of social brain networks (including the medial prefrontal lobe, temporoparietal symphysis, amygdala, etc.) is closely related to social impairment. On the other hand, hundreds of genetic variants have been found to be associated with autism susceptibility, but a single gene often has limited contributions, and the genetic mechanism of autism is highly complex. Therefore, how to associate genes, neural circuits, and cognitive behaviors across scales has become an urgent scientific problem to be solved. In addition, the rapid development of artificial intelligence technology provides new ideas for auxiliary diagnosis and behavioral intervention of autism. In recent years, a large number of studies have attempted to use machine learning models to identify autism markers from brain images or behavioral videos. In particular, multimodal deep learning fusion technology is being used to extract complex correlation features in order to improve diagnostic accuracy. However, most of the current AI diagnostic models are still limited to a single level of data-driven, lacking an explanation of the nature of cognitive deficits in autism, and belong to a typical "black box" model. Internationally, there is a growing need for explainable and controllable AI-powered diagnostic tools. To sum up, the development of a new theoretical and technical framework that can connect multiple levels of genes, brains and cognition is not only an international frontier trend, but also a strategic need at the national level.

Originality and theoretical value of the DIKWP model: In the face of the above challenges, Professor Duan Yucong proposed the DIKWP model with independent intellectual property rights, which provides a new paradigm for the cognitive modeling of complex intelligent systems. "DIKWP" stands for Data-Information-Knowledge-Wisdom-Purpose, which is based on the traditional DIKW (pyramid model: data, information, knowledge, wisdom) and adds an original "intention/purpose" layer. The model adopts a mesh interaction architecture, which breaks the limitation of linear hierarchy, so that the semantics of each level can be fed back and updated in both directions. The core idea of this design is that by explicitly introducing purposiveness into the cognitive system, machine intelligence can have a rudimentary form of "self-awareness", capable of self-monitoring, self-reflection, and regulation of cognitive processes. The DIKWP model is an academic milestone and is regarded as an innovative way to solve the "black box" problem of artificial intelligence and improve the interpretability and controllability of AI systems. "The DIKWP model builds a common cognitive language for human institutions, so that every step of the AI decision can be traced, interpreted, and understood by humans. By embedding the key layer of 'purpose' into the model, we are not only making AI smarter, but also ensuring that it remains in service of human values and security needs." “。 At present, the model and its related artificial consciousness theory have obtained 114 domestic and foreign invention patents (including 15 PCT international patents). The DIKWP model has shown great potential in the fields of cognitive computing and artificial consciousness systems, such as building an interpretable cognitive operating system, and dismantling large model reasoning into five monitorable links: data, information, knowledge, wisdom, and intention, so as to ensure that each step of AI output can be traced. These achievements have attracted wide attention around the world, making DIKWP a new vane for AI explainability and security research. It can be seen that DIKWP represents the cutting-edge theory of independent innovation in China, and has the world's leading originality in intelligent cognitive modeling.

New significance of the DIKWP model for autism research: The introduction of the DIKWP model into the study of autism is expected to reconstruct and expand the traditional research path of autism, and provide a new perspective for revealing the mechanism of social cognitive deficits in autism. The core obstacle of people with autism is the perception and understanding of social information, that is, the inability to effectively extract data/information from other people's behaviors and rise to social knowledgeand then infer the intentions and purposes of others (which is the lack of "theoretical psychology" (ToM) ability). This is highly consistent with the deficits of functions at all levels of the DIKWP model: children with autism often show abnormal perception of social stimuli (such as facial expressions and eye contact), and it is difficult to obtain correct information from sensory data in time (such as identifying emotional expressions); They also have barriers to semantic comprehension and situational reasoning (knowledge to intelligence), and are unable to integrate scattered information into knowledge or strategies for social situations; Most notably, they have difficulty guessing the intentions/purposes of others, i.e., deficiencies in social intelligence and mental understanding, which is the ability represented by "Purpose" at the top level of the DIKWP model. Therefore, the DIKWP model provides a systematic framework for analyzing why autistic patients have functional deviations at different cognitive levels: we can treat autism as a cross-level cognitive processing disorder coupled neural circuit abnormality, and use DIKWP to characterize this coupling relationship. This convergence of theories is expected to answer several key questions in autism research, such as: How are abnormalities in sensory input progressively amplified into high-level social cognitive deficits? How does damage to specific circuits of the brain, such as those responsible for facial recognition or empathy, affect the transformation of information into knowledge and intelligence? How does the lack of social motivation (lack of social intention) manifest itself at the brain and behavioral level? These are all scientific topics that DIKWP can focus on.

Importance of this project: Based on the above background, the theme of this project is "Research on the Neural Mechanism and Gene Editing Therapy of Autism Integrating DIKWP Model**", aiming to create a new research paradigm of autism guided by artificial consciousness. This will comprehensively enhance China's original innovation ability in the field of basic research and clinical intervention for autism, and open up a new path for the integration of brain science and artificial intelligence. The significance of the project is embodied in the following aspects:** (1) Theoretical innovation: Introduce the DIKWP artificial consciousness model into the study of the social cognitive mechanism of autism, fill the gap of cross-level integration theory, and enrich the application theory of artificial intelligence in brain diseases; (2) Technological breakthroughs: build a technical chain of "perception-knowledge-expression" to realize the closed-loop from the patient's perception layer (neural signals, behavioral data) to the knowledge layer (cognitive state modeling) to the expression layer (clinical decision-making, intervention plan), and improve the intelligent and accurate level of autism diagnosis and treatment; (3) Practical value: to develop interpretable multimodal AI-assisted diagnosis systems and novel gene editing treatment strategies to meet the urgent clinical needs for early objective diagnosis and effective intervention, and improve the prognosis and quality of life of children with autism; (4) Strategic effect: To create a research platform for artificial awareness of autism and a new paradigm of intelligent diagnosis and treatment, to help China produce a number of landmark achievements with international influence in the field of brain-intelligence intersection, to serve national strategies such as "Healthy China 2030", and to provide reference for the research of other neurodevelopmental diseases, which has significant social and economic benefits.

To sum up, this project closely combines the national strategic needs, international frontier trends and the original theoretical advantages of Professor Duan Yucong's team in our university, and is expected to achieve a breakthrough in the field of autism mechanism and treatment, and promote China to enter the global leading ranks in the new generation of artificial intelligence + brain science integration research.

2. Research objectives and overall technical roadmap

Overall Objectives: The overall goal of this project is to construct a cross-level research framework that integrates the DIKWP model to comprehensively and deeply study the pathogenesis and potential treatment pathways of autism from neural circuits, cognitive processes to interventions. By introducing the hierarchical cognitive architecture of the DIKWP model, we hope to elucidate the coupling relationship between core brain circuit abnormalities and information-knowledge processing process disorders in patients with autism**, develop a new multimodal diagnosis technology based on artificial consciousness, and explore the feasibility of gene editing in autism intervention. Specifically, the project seeks to achieve the following key objectives:**

Objective 1: To uncover the "perception-cognition" coupling law of the neural mechanism of autism. The DIKWP model was used to model the abnormal connection of multiple brain regions in patients with autism, analyze how impaired sensory data/information processing leads to defects in the formation of high-level knowledge/wisdom, and elucidate the neural circuit basis of social cognitive disorders in autism. The focus is to find the correspondence between abnormalities in the brain's core circuits (e.g., social brain networks, mirror neuron systems, etc.) and cognitive deficits (e.g., emotional perception, language comprehension, intentional reasoning disorders).

Objective 2: Develop a multimodal AI diagnosis system **based on DIKWP artificial consciousness. **Design an artificial intelligence diagnostic model that integrates the hierarchical semantics of "data-information-knowledge-intelligence-intention", and integrates multimodal information such as clinical assessment, behavioral video, brain imaging and genetic data. Realize the early automatic identification of autism, make the diagnosis process transparent and explainable, and trace the basis of AI decision-making, so as to improve the accuracy and credibility of diagnosis and make up for the shortcomings of traditional manual assessment.

Goal 3: To explore new targets and strategies for gene editing therapy for autism. Based on the in-depth understanding of the cognitive circuit of autism and the simulation and prediction of the artificial consciousness model, the possible key genetic/molecular targets were screened out using CRISPR/Cas9and other gene editing technologies have been validated in cells and animal models. To evaluate the effect of correcting these genetic defects on neural circuit activity and behavioral phenotypes to provide a potential underlying treatment for autism. Particular attention is paid to molecules that regulate the flow of information at various layers of the DIKWP model, such as signaling pathways that affect synaptic plasticity (data/information processing level) or social motivation and cognitive flexibility (intelligence/intention level).

**Goal 4: To build an integrated platform for autism assessment and intervention driven by artificial awareness. **The above research results are integrated and applied to clinical practice, and an intelligent system including AI diagnosis, cognitive function assessment and personalized intervention recommendations is built. The system was deployed in actual medical and rehabilitation scenarios to evaluate cognitive abnormalities and monitor the effect of intervention in children with autism, and verify its effectiveness and practicability. Finally, a new paradigm of artificial intelligence diagnosis and treatment of autism that can be promoted will be formed, laying the foundation for its application nationwide in the future.

Overall technical route: In order to achieve the above goals, this project adopts the overall technical route of "traditional path + DIKWP empowerment", that is, the classic "three links" in autism research**: neural circuit mechanism**, AI-assisted diagnosis, and gene therapy exploration——The DIKWP model is introduced as a guiding framework throughout the whole process, and a "perception-knowledge-expression" connected technology chain from basic mechanism to clinical transformation is established. Figure 1 (omitted) illustrates the overall scheme of the project.

Specifically, we divide our work into four levels that are connected to each other: (1) Perception layer (data-information layer): focus on the collection of physiological and behavioral data of patients with autism, including brain imaging, EEG, gene sequencing, cognitive testing, and social behavior video. Through signal processing and feature extraction, massive raw data is transformed into meaningful information representations, such as brain functional connectivity matrices, behavioral coding indicators, mutant gene lists, etc. (2) Knowledge layer (knowledge-wisdom layer): Based on the DIKWP cognitive architecture, the aforementioned information is fused and high-level semantic modeling is carried out to construct a knowledge graph or cognitive computing model that reflects the cognitive state of autism. Here, we use artificial consciousness algorithms to further abstract information into knowledge (e.g., identify specific deficit patterns in the patient's emotion recognition, language comprehension, etc.), and make inferences at the intelligence layer (e.g., synthesize multiple defects to infer the overall diagnostic conclusion). The two-way feedback properties of the DIKWP model will also be used to simulate cognitive adjustment processes within patients, such as predicting how high-level cognition adjusts when underlying sensory input changes. (3) Expression layer (intent/application layer): This is the output link of the technology chain, which involves applying the results of cognitive models to the expression of diagnostic decisions and intervention strategies. On the one hand, the developed AI diagnostic system will give an interpretable diagnosis report, including various layers of basis (data anomalies, information characteristics, knowledge reasoning, intelligent decision-making, and potential intention analysis) for clinicians to refer to for decision-making. On the other hand, in terms of intervention, this layer will output potential therapeutic targets and suggestions for individuals (for example, the abnormality of a certain gene function has a significant impact on their ability to reason about their intentions, it is recommended to consider corresponding drugs/gene therapies), and then act on the upper layer research as feedback to verify the model prediction. Through the close connection of these three layers, we have formed a closed-loop research route: from perceptual acquisition of data, to cognitive modeling at the knowledge layer, to the application of expression results, and in turn to test and optimize the model.

It is worth emphasizing that the DIKWP model is not simply attached to the above processes, but is deeply integrated into each link, leading the research ideas and technical realization: in the study of neural mechanisms, it helps us design experimental paradigms across brain regions and functional levels; In the AI model, it is directly used as the architecture blueprint to determine the division and interaction of algorithm modules. In genetic and clinical applications, it guides us to screen targets and design evaluation indicators from the perspective of cognitive function. Therefore, DIKWP provides a unified semantic coordinate system and logical thread to ensure that each module has its own focus but is interconnected, and ultimately serves the overall goal of revealing and intervening in autism.

Innovative approaches versus traditional research: Traditional approaches to autism research tend to work in silos—neuroscientists focus on brain abnormalities, AI experts develop diagnostic tools, and geneticists search for genes, but these efforts lack a unified framework. In contrast, the technical route after the introduction of the DIKWP model in this project has three innovative features: one is hierarchical connectivity, we use the three stages of perception-knowledge-expression to connect the basic and clinical, so that abnormal brain activity can explain its behavioral consequences through the cognitive model, and can be directly used to guide the selection of intervention targets. The second is model-driven, the DIKWP model provides an embedded purposeful agent architecture, so that AI diagnosis is no longer an empirical pattern classification, but an active intelligence with human-like cognitive processes, sustainable learning and self-optimization. The third is closed-loop validation, in which gene editing experiments provide validation feedback for model prediction, and the model can be adaptively adjusted according to the experimental results (such as modifying the knowledge graph), forming a combination of data-driven and mechanism-driven cycles. This new paradigm will greatly improve the explanatory power and relevance of research: for every brain connectivity abnormality found, we can immediately assess how it disrupts the flow of information; For each proposed intervention, we can predict its impact on cognitive circuitry through the model. This ensures that the results are both theoretical and practical.

To sum up, the overall technical route of this project is guided by the DIKWP artificial consciousness model, and runs through all links of "mechanism research-diagnostic model-treatment exploration-clinical validation", forming an interdisciplinary cross-integration, evidence-based iterative optimization model. This route is closely related to the pain points of autism research, and will provide solid support for breaking through the bottleneck of autism cognitive mechanism and intervention.

3. Research content and key modules

Focusing on the above goals and technical routes, this project has set up four research content modules, which correspond to different levels of autism research. In each module, we will integrate the ideas and methods of the DIKWP model to form several key technological breakthroughs. The modules and their main research contents are as follows:

3.1 Modeling of autism neural mechanisms (perceptual layer & information-knowledge mapping)

Overview: This module focuses on the modeling of brain structure and functional abnormalities in autism, and aims to reconstruct the mapping relationship between multi-brain region connectivity and cognitive deficits in combination with the DIKWP model. We will systematically collect multimodal brain data from patients with autism and typical developmental populations, including functional magnetic resonance imaging (fMRI), resting-state EEG/MEG, diffusion tensor imaging (DTI), etc., to quantify brain network topological differences and information transfer efficiency. At the same time, the DIKWP model is introduced to link the data of these neural "perception layers" with higher-level "knowledge/intelligence" functions to explain how specific brain circuit abnormalities lead to cognitive impairment.

Key scientific questions: The neural mechanism of autism can be summarized as the loss of information between the underlying processing processes such as sensation, attention, memory, and emotion, and the high-level functions such as social cognition. We will focus on the following scientific questions:

*(1) How does abnormal functional connectivity in multiple brain regions affect information processing? *Autism research has shown that long-term brain connections tend to be weakened and local connections may be too strong, the so-called "brain malconnection" theory. We intend to examine the patterns of impaired information flow transmission between different regions in social cognition-related networks (e.g., default mode network (DMN) and social cognitive network, and how this impairment corresponds to the failure of social information integration. From the perspective of DIKWP, we hypothesize that autistic people may not be able to effectively transmit key signals to higher-order regions due to abnormal sensory filtering or attentional selection mechanisms at the "data → information" level, resulting in incomplete representation at the subsequent "knowledge" level. For example, fMRI was used to construct a whole-brain functional connectivity map, and the information mediation degree of each node was measured in combination with Graph theory, so as to find the information convergence hub that was significantly reduced in the autism group, so as to locate the key loops that led to the interruption of information flow.

*(2) How are cognitive deficits represented in brain networks? *Cognitive deficits in autism include social perception (facial expressions and eye perception), emotion recognition, language comprehension, and executive function(planning, flexibility) and other aspects. We will design a series of cognitive tasks (such as facial emotion recognition tasks, virtual social interaction tasks, etc.) combined with fMRI to observe the differences in brain activation and changes in network collaboration patterns in patients with autism when completing these tasks. With the DIKWP model, we will map different task requirements to cognitive levels: for example, face emotion recognition involves more processing of "information → knowledge" (from visual features to semantic emotional concepts), while intentional reasoning tasks involve more processing of "knowledge → intelligence/intention" (inferring the purpose of others from known cues). Comparing the similarities and differences between the brain activity pathways of autism and control groups under these tasks can reveal the dysfunction of autism at specific DIKWP levels. For example, if the prefrontal-temporoparietal node synergy is significantly reduced in the autistic group when reasoning about the intentions of others, this will confirm the "intelligence/intention" layer processing disorder and can be related to the degree of theoretical psychological deficits assessed by clinical assessment.

*and (3) multi-scale unified characterization of neural circuit abnormalities. *Abnormalities in the brain can be described at multiple scales (molecular/cellular, local circuits, whole-brain networks). We plan to construct a multi-scale brain connection-cognitive map: the upper nodes represent cognitive functional units (such as emotion recognition module and language comprehension module), the lower nodes represent anatomical structures (brain regions and neural circuits), and the two layers are connected by functional mapping (quantitative calculation of mapping strength based on experimental data). This two-layer map is similar to the hierarchy of the DIKWP model, and can be regarded as the implementation of the DIKWP framework in the neural dimension, where the lower-level biological network realizes the upper-layer cognitive function. By comparing the topological differences in autism and control profiles, we will identify autism-specific key connectivity deletions or abnormal connectivity increases. For example, if it is found that the functional node of "eye contact perception" in the autism atlas is weakly connected with the subordinate occipital visual cortex node, it indicates that the transmission of visual perception to social cognition is impaired (the disconnection between the perceptual layer and the knowledge layer). This map can also be used in conjunction with the genetic module to indicate which autism risk genes each edge may be subject to (detailed in Section 3.3) to provide guidance for subsequent interventions.

Research Methods and Technical Roadmap: In order to achieve the above goals, this module will comprehensively use a variety of brain imaging analysis and computational modeling techniques

In terms of data acquisition, we cooperated with clinical hospitals to recruit a large sample of autistic children and matched healthy controls, and collected high temporal resolution (EEG/MEG) and high spatial resolution (fMRI, DTI) data, as well as neuropsychological evaluation results. Data acquisition for children with autism will focus on reducing motor artifacts and improving task cooperation, such as using a gamified fMRI task paradigm to improve child cooperation.

In terms of signal analysis, functional connection analysis, independent component analysis (ICA), spectrum analysis and other methods were used to extract the neural features of each modality. For example, the resting state and task state functional connectivity matrices were extracted from fMRI, the EEG power spectrum and phase coupling indexes were extracted from EEG, and the white matter fiber bundle connection map was reconstructed from DTI.

In terms of DIKWP integration, a cognitive graph construction algorithm was developed: each subject was represented as a multi-layer network by combining brain connectivity features and cognitive test/task performance. Among them, the underlying network nodes are brain regions and pathways (according to anatomical or functional distinctions, such as visual, auditory, social brain regions, etc.), and the upper network nodes are cognitive function units (defined according to test items, such as emotion recognition, working memory, language comprehension, etc.). The upper and lower layers are connected by edges, and the weights are represented by the contribution of the brain region to the corresponding cognitive function (learning from the data can be achieved through machine learning regression models or structural equation models).

Group comparison and model discovery: The cognitive graphs of the autism group and the control group were compared, the differences were quantified by graph theory indicators (clustering coefficient, between-centrality, etc.), and the community detection algorithm was used to find abnormal connection clusters. It is expected to find a decoupling of specific modules in the autism group, such as loose connections within the social cognitive module, or weak connections between the sensory-cognitive layers.

DIKWP simulation verification: A computer simulation model (such as a neural network model) is constructed based on the obtained graphs to simulate the transmission process of information from the sensory layer through the knowledge layer to the decision-making layer. The simulation model of autistic people was simulated by intervention (such as strengthening a virtual connection or introducing noise at a specific node) to observe the impact on the output behavior, and theoretically verified the causal relationship of how a certain circuit abnormality leads to cognitive deficits. Such simulations provide a hypothetical basis for subsequent genetic or neuromodulation experiments.

Expected Results: This module will produce a series of high-resolution brain network maps and cognitive mapping models on the neural mechanisms of autism. It is expected to theoretically put forward new evidence for the hypothesis of "autism information processing circuit loss", and find several key neural pathways, whose dysfunction directly leads to the blockage of the transformation of information into knowledge in the DIKWP hierarchy. In terms of data, the world's first multi-scale brain connection-cognitive database for autism integrating the concept of DIKWP was constructed to realize the full observation and description of the autism brain network. In terms of methodology, a new method of integrating brain network and knowledge graph has been developed, which sets a model for the study of cross-level brain disease mechanisms. The results of this module will not only serve the follow-up AI diagnosis and gene intervention module (providing key features and targets), but will also be published in the form of papers to enhance China's international influence in the field of cognitive neuroscience.

3.2 Multimodal AI-assisted diagnosis (Knowledge Layer & Intelligence-Purpose Artificial Consciousness Diagnosis Network)

Overview: This module focuses on the construction of a multimodal artificial intelligence autism diagnosis system, making full use of the information-knowledge-intelligence hierarchical structure of the DIKWP model, and developing an artificial consciousness diagnosis network with stronger interpretability and flexible adaptability. The system will integrate the patient's multi-source data (including clinical questionnaires, behavioral videos, voice patterns, eye movements, EEG/fMRI features, genetic information, etc.), and automatically determine whether an individual belongs to the autism spectrum and give a basis for diagnosis through hierarchical feature extraction and cognitive reasoning. This module seeks to break through the bottleneck of existing AI diagnosis, so that model decision-making is no longer an incomprehensible black box, but can simulate human cognitive processes, so that doctors and parents can be convinced of their conclusions.

Key innovations and difficulties: The difficulty of multimodal AI diagnosis lies in the fact that different types of data have different scales, non-linear correlations, and limited and heterogeneous sample data. The innovation of this module lies in the introduction of the DIKWP artificial consciousness architecture, which disassembles the diagnostic process into step-by-step reasoning similar to human cognition: from the perception of raw data, to the understanding of intermediate semantic information, to high-level knowledge and intelligent decision-making, and finally to the calibration results combined with diagnostic intent. Specific innovations include:

(1) Hierarchical feature extraction and semantic mapping: We will design a multi-level deep learning model, and the first layer will process the raw data of each modality to realize the underlying feature extraction (corresponding to the data/information layer of DIKWP). For example, convolutional neural networks extract micro-expression features in facial videos, speech models extract speech tone features, and graph neural networks process brain connection matrix features. The second-layer model maps these extracted information features to intermediate semantic representations, such as fusing video and audio features to judge the quality of social interactions, and correlating EEG features to cognitive states (attention levels, mood swings, etc.). This layer is equivalent to the knowledge layer construction of DIKWP, we will refer to the knowledge graph or expert experience, assign interpretable semantic labels to the middle nodes of the model (such as "low frequency of eye contact", "single intonation", etc.), and track which original features contribute the most to these intermediate conclusions through attention mechanism or saliency analysis techniques to enhance interpretability.

(2) Artificial Consciousness Decision-making Network: At the senior level, we will build an artificial consciousness decision-making module with metacognitive capabilities. This module is equivalent to the intelligence and purpose layer of the DIKWP model, which is responsible for synthesizing various intermediate semantic evidence, making diagnostic inferences, and conducting self-assessment and adjustment. We will draw on the "dual circulation" architecture proposed by Professor Duan Yucong: that is, in addition to the basic cognitive process (feature → discrimination), a metacognitive cycle is added. The basic process outputs a preliminary diagnosis (e.g., autism score), and the metacognitive loop monitors whether this output is consistent with the purpose built into the system. The "purpose" here refers to the human expert diagnostic criteria and social value that the model internalizes at the time of training: for example, avoiding over-diagnosis or missing a diagnosis. By checking the credibility of each step in the inference chain (data→ information→ knowledge → decision-making), the metacognitive module can trigger self-adjustment mechanisms if a step is found to be highly uncertain or unfit for purpose (e.g., a key sign is diagnosed without being explained): for example, requesting more data, adjusting the weight of a feature, or suggesting a referral to an expert for re-evaluation by giving an "uncertain" conclusion. This mechanism simulates the reflection process in human diagnosis, making AI systems more robust and safe. Our model will use reinforcement learning or generative adversarial training to allow the metacognitive module to learn to avoid bias while maintaining accurate diagnosis, and truly realize "intelligent self-knowledge".

(3) Interpretable presentation of diagnostic results: The final output of the system is not only a simple "positive/negative" judgment, but also a detailed interpretable report. We will design a natural language generation module that transforms the reasoning process inside the model into a human-readable report. The report may include: primary data characteristics (e.g., "the child spent only 30% of the normal mean fixation on the face"), intermediate inferences ("this suggests a significant lack of social attention"), a comprehensive analysis ("combined with the monotony of the voice and repetitive behavior, indicating a social communication disorder"), as well as confidence and possible sources of error. The report also provides recommendations for next steps (e.g., "Genetic testing is recommended to check for Fragile X syndrome" or "Social training is recommended") that are rooted in the model's "intent" layer knowledge base (trained from expert experience). Through this intelligent assistant-like interpretation report, doctors are able to understand the AI decision-making logic and adjust the diagnostic strategy accordingly. Parents can also get clear feedback on what their child's problems are and why, and improve acceptance of the diagnosis. The introduction of explainability will greatly promote the reliable application of AI diagnosis in clinical scenarios.

Research Methods and Implementation Steps: The implementation of this module will go through three stages: model design, training validation, and clinical testing

Model design: Based on repeated discussions with clinical experts, the list of modalities and features that need to be integrated into the diagnostic system is determined, and the model module structure is divided based on DIKWP. Based on the results of module 3.1 mentioned above, the neural features (such as specific brain network connectivity indicators) and cognitive and behavioral features that are most valuable for diagnosis were extracted and included in the model input. Then, a modular design method is used to build a multi-branch neural network: each data mode corresponds to a network branch to extract low-level features, and then perform feature fusion and semantic mapping in the middle layer. In order to add a metacognitive unit to the high-level decision-making part, a mechanism needs to be designed to measure the deviation between the decision and the purpose, and we consider using a probabilistic graph model or a Bayesian network to model the inference chain to calculate the uncertainty.

Model training: Leverage large-scale data for model training and tuning. The data sources include clinical data collected by the project itself, as well as open data obtained in cooperation with domestic and foreign autism databases. In the training process, in order to solve the problems of multimodal fusion and small shots, we will introduce strategies such as transfer learning and federated learning: for example, each branch network is pre-trained on the relevant tasks of large samples to master the preliminary feature extraction ability, and then the joint training model is fine-tuned on the data of this project. At the same time, we should pay attention to interpretable regularization, and add to the loss function to encourage the accurate prediction of interpretable semantics by the middle layer of the model (such as predicting clinical scale scores at the same time). The model structure is continuously improved by using the data annotated by experts to verify whether the interpretation of the model output is in line with medical knowledge (e.g., checking whether the area of interest of the Attention mechanism is a face).

Model Validation and Clinical Testing: In offline validation, we will use multiple evaluation indicators: diagnostic accuracy (including sensitivity and specificity), model stability (performance in different data cases), and interpretability scores (Clinical experts are invited to score according to the report) and so on, and conduct a comprehensive evaluation of the model. This was then piloted in a real-world clinical setting: in collaboration with the autism diagnosis centers of partner hospitals, our AI systems were embedded in their diagnostic processes. Based on the routine assessment, the doctor refers to the AI report to record the consistency of the important findings provided by the AI with the final diagnosis, and whether the AI helped the doctor find clues that were overlooked. Through prospective trials of a certain scale of cases, the utility and reliability of the system in the real environment were verified, and feedback from doctors was collected to iteratively improve the model.

Expected Results: The direct result of this module will be a multimodal diagnostic AI system for autism with independent intellectual property rights (presented in the form of software), and its performance is expected to exceed the level reported in existing studies. In terms of technical indicators, it is expected that the diagnostic accuracy rate can reach more than 90% in the test set we collect, which is significantly higher than that of traditional methods. In particular, early screening of young children under 2 years of age is expected to increase sensitivity by at least 20% compared with current questionnaire methods. More importantly, we have improved the explainability of the quantitative display system: for example, the AI report has been proved to be more than 80% consistent with human professional understanding through expert scoring, which is significantly better than the black box model without the DIKWP architecture. This achievement will be published in the form of a paper or invention patent, which will establish our leading position in the field of AI-assisted autism diagnosis and promote the standardized application of AI in the field of mental health.

In addition, as an application demonstration of artificial consciousness theory, the system will verify the effectiveness of the DIKWP model in practical complex tasks, which has important academic value. The diagnostic reports and intermediate analysis data generated by the system will also be fed back into the basic research module to help us better understand the characteristic patterns and underlying mechanisms of autism. In conclusion, this module will produce a set of practical and advanced intelligent diagnostic tools, which will provide the possibility for the early detection and intervention of autism, and alleviate the dilemma of insufficient clinical diagnosis resources to a certain extent.

3.3 Gene Editing and Therapeutic Validation (Intelligence Layer &; Gene-Loop-Cognitive Feedback)

Overview: This module looks at the molecular and genetic aspects of autism and aims to explore the feasibility of using gene editing technology as an intervention. Based on the aforementioned studies of neural mechanisms and cognitive models, we will identify key genes/molecules that are highly associated with core deficits in autismand verify its causal effects through experiments at the cellular and animal levels. Furthermore, cutting-edge gene editing tools such as CRISPR/Cas9 were used to repair or regulate these genes to see if they could reverse the related neural circuit abnormalities and behavioral phenotypes. This module will transform the mechanistic cognition obtained from the artificial consciousness model into potential treatment strategies, realize the closed loop from theoretical prediction to experimental verification, and promote the extension of autism intervention methods from behavioral training to biomedical level.

Rationale: Genetics studies of autism have identified hundreds of risk genes and several monogenic syndromes (e.g., Rett syndrome/MECP2, fragile X syndrome/FMR1, tuberous sclerosis). /TSC1/2, etc.). Most of these genes are involved in key pathways such as synaptic development, neural connectivity, and gene expression regulation, suggesting that autism is largely a synaptic disease or brain junction disease. With the maturity of gene editing technology, the treatment of brain diseases with single gene defects is emerging, such as gene therapy for ASD-associated Angelman syndrome (UBE3A gene inactivation). Therefore, it is reasonable to explore: for those genes that are highly related to social deficits in autism, can they be edited to correct the symptoms to alleviate the symptoms? The unique role of the DIKWP model here is to provide a basis for screening and interpretation: the model can reveal how certain genes disrupt information processing by affecting neural circuits. For example, if the model predicts that the loss of function of a synaptic protein gene (e.g., SHANK3) will result in "inability to upload information from the perceptual layer to the knowledge layer" (e.g., due to decreased neural signaling efficiency), then SHANK3 becomes an attractive target for intervention. Correspondingly, we will focus on genes that play a key role in the DIKWP process (e.g., molecular pathways involved in social memory, motivational drive: oxytocin, 5-HT; or genes involved in the balance of neural excitation/inhibition: SCN2A, GABA receptors, etc.).

Main research steps and contents:

(1) Screening candidate gene targets: Utilize the results of existing large-scale gene association studies and whole-exome/whole-genome sequencing data from our own samples to create a list of genes that are significantly associated with the autism phenotype (e.g., the top 50 genes with the highest autism risk scores). Combining the findings of modules 3.1 and 3.2, we mapped these genes to cognitive maps and brain networks to look for links where they may influence. For example, if a gene is highly expressed in the prefrontal lobe and is involved in synaptic plasticity, it is speculated that it affects the flexible decision-making function of the "knowledge-wisdom" layer. If a gene affects cerebellar development, it may be linked to sensory-motor coordination, which has an impact on the processing of "data → information". We will validate these inferences through literature review and bioinformatics analysis (gene function annotation, pathway enrichment analysis). Finally, a small number of key genes were selected as intervention candidates, and the following types were preferred: (1) single-gene mutations were clear: such as SHANK3 (Phelan-McDermid syndrome) or MECP2 (Rett), whose mutations lead to autism-like manifestations, and substitution/repair can theoretically play a role; (2) High-frequency risk genes: such as CHD8, SCN2A, etc., which have been repeatedly verified by multiple studies; (3) Druggable/editable targets: gene products are enzymes or receptors, which can be regulated by pharmacological means or gene editing methods, such as mTOR pathway genes, OXTR (oxytocin receptor), etc.

(2) Cell-level functional validation: For the key genes screened, we will first verify their effects on neuronal function and DIKWP cognitive processes in an in vitro model. On the one hand, the 2D culture model of brain-like organoids or neurons obtained from the differentiation of patient-induced pluripotent stem cells (iPSCs) was used to record the abnormalities of the cells carrying candidate gene mutations, such as synaptic release, growth cone movement, and network electrical activity. On the other hand, CRISPR/Cas9 gene editing is used to perform correction or knockout experiments at the cellular level: for example, the patient's mutated gene is corrected to a normal sequence to observe whether normal neuronal function is restored; or introduce the mutation into healthy control cells to see if the autism-related abnormalities are reproduced. Calcium imaging, patch-clamp, electrophysiological array and other techniques were used to measure the differences in neural activity patterns before and after editing. For example, CRISPR activation of normal allele expression in mouse neurons has been studied to compensate for the loss of function of SCN2A monoallelic inactivation. Similarly, we expect to see editing corrections of key genes that enhance neuronal excitability or synaptic maturation, thereby partially restoring the normal rhythm of "data/information" processing, laying the groundwork for further validation in animals.

(3) Gene therapy trials in animal models: Based on the positive results of cell validation, we will select appropriate animal models for higher-level therapy evaluation. For each candidate gene, there is usually a corresponding genetically engineered model mouse in the literature (e.g., Shank3 knockout mice exhibit social deficits; CNTNAP2 knock rats have autistic behavior, etc.). We will introduce these mouse models and confirm that their phenotypes are consistent with our expectations (e.g., decreased social interactions, increased repetitive behaviors, etc.) through behavioral testing. Subsequently, gene editing therapy strategies are designed to intervene in model mice, such as using AAV viral vectors to deliver the CRISPR system to specific regions of the brain to knock out genes that inhibit overactivity, or activate silencing alleles. Attention needs to be paid to controlling the specificity and safety of editing, and we will select validated high-fidelity Cas9 and suitable gRNA to avoid off-target effects. After the intervention, the mice were evaluated for improvement in core behaviors (time spent on social interaction, ultrasound vocal communication, number of stereotyped movements), as well as changes in brain electrophysiology and connectome. For example, if gene therapy is administered to Shank3-deficient mice, we expect to observe enhanced hippocampus-prefrontal synaptic transmission, increased social interest, etc. We also plan to use chemogenetics**/optogenetics to temporarily modulate the corresponding circuits to verify that the behavioral changes are indeed caused by the restoration of gene function—which is equivalent to a validation of the DIKWP model: when the data-information transfer of a certain loop is restored**, the corresponding cognitive function should be restored. If the results of animal experiments show that gene editing can significantly improve the symptoms of the model, it will lay the foundation for clinical human gene therapy in the future.

(4) DIKWP model feedback and mechanism interpretation: After obtaining the experimental data, we will use the DIKWP model to interpret the results before and after the intervention. For example, in animal experiments, we can collect brain functional imaging and behavioral data from mice before and after gene therapy, build corresponding cognitive maps, and compare changes in information flow indicators. If it is found that some key information flows (e.g., sensory input to the social cognition module) are restored, then this is consistent with the prediction of the DIKWP model, which strengthens the correctness of the model; Conversely, if there are unforeseen outcomes, it can prompt us to revise the model's hypotheses about the mechanism of action of the gene. In this way, the model and experiment are mutually validated and improved, and finally a more complete understanding of the biological mechanism of autism is formed. For example, we may propose that "gene X fails to effectively translate social information into knowledge-layer representations in the autistic brain by influencing the strength of the connection between Y and Z in the brain; Gene therapy enhances Y-Z connections, repairs information flow, and improves social behavior", a new mechanistic insight that is both a triumph for the DIKWP model and a significant contribution to the understanding of autism pathology.

Expected Result: This module will achieve the following outcomes:

Scientific discovery level: identify 1-2 molecular pathways that have a decisive impact on social cognitive impairment in autism, and verify their causal effects through gene editing. For example, we are expected to show that "restoring the function of the gene SHANK3 can reverse some social deficits in autism" or "inhibiting the overactive mTOR pathway can improve repetitive behaviors in autism models", these findings will be published in high-level journals and attract academic attention.

Technological breakthrough: forming a gene editing intervention paradigm for autism. While gene therapy is currently focused on monogenic genetic diseases, our work will be expanded to the complex disease autism to be the first to provide proof-of-concept data. Relevant technical achievements (such as AAV-CRISPR vector construction methods for specific genes, mouse behavior improvement data) can be patented, laying the foundation for future industrialization.

Theoretical model level: Through the combination of experiment and model, the coverage of the DIKWP artificial consciousness model on the biological level is greatly enriched. We will update the parameters of the model to quantitatively simulate the influence of genes on cognitive processes, and form a "gene-brain-cognition" trinity model of autism mechanism. The model itself is an innovation that can be used as a digital twin to test the effects of various intervention strategies in a virtual environment and accelerate the development process.

In conclusion, the results of this module will mark a key step in our exploration of autism treatment, expanding from relying solely on behavioral/educational interventions to biomedical correction, providing hope for a fundamental cure for autism. At the same time, the new theories and technologies generated by the project will also have a demonstration effect on the research and treatment of other neurodevelopmental disorders (intellectual disability, attention deficit, etc.).

3.4 Clinical Research and System Convergence (Purpose Layer & Platform Integration & Application Demonstration)

Overview: This module aims to integrate and apply the results of the above studies to the actual clinical setting, construct an artificial awareness-driven autism assessment and intervention system, and verify its effectiveness and generalizability through clinical trials. Specifically, the AI diagnostic system developed in 3.2 will be connected to the existing diagnosis and treatment process in the hospital, and clinicians will be trained to use and evaluate its diagnostic performance improvement, the potential therapeutic targets and intervention methods screened in 3.3 will be integrated into the comprehensive clinical autism intervention plan to explore a new model of personalized treatment, and the artificial awareness research platform for autism will be established to integrate data, models and applications, so as to realize the long-term sharing and update iteration of project results. Eventually, this module will output a set of practical clinical solutions and platforms, paving the way for the promotion and transformation of results.

Main Research and Work Tasks:

(1) Clinical deployment and verification of the intelligent diagnosis system: The child psychology department/rehabilitation department of the cooperative hospital was selected as a pilot, and the multimodal AI diagnosis system completed by module 3.2 was deployed to the real clinic environment. First, medical staff were trained to master the use of the system, including data collection (such as entering children's behavior videos, importing EEG results, etc.), AI analysis triggering, report interpretation, etc. Then, when receiving new patients in the outpatient clinic, an AI system is introduced to participate in the evaluation: the system analyzes the patient's data and gives predictions and reports on the possibility of autism, and then the clinician makes a comprehensive judgment together with the routine diagnosis. We will collect data from all children who come to the clinic over a period of time, and compare the joint diagnosis conclusions of AI+ doctors with the conclusions of traditional doctors only and the results of follow-up visits, and calculate the auxiliary value of AI. For example, whether AI improves the detection rate of mild autism and reduces missed and misdiagnosed cases. In controversial cases, an independent expert assessment is organized to determine the reasonableness of the AI recommendations. In addition to diagnostic accuracy, physicians' satisfaction and compliance with the system (through questionnaires) are also evaluated, the user-friendliness of the system is understood, and the interpretation report helps clinical decision-making. If it is found that some modules of the system are not suitable for clinical practice (such as data acquisition takes too long or the report is too professional and difficult to understand), feedback to the R&D team for optimization and improvement in time. After the success of the pilot, develop a standardized clinical application process and user manual to prepare for a large-scale rollout.

(2) Application of artificial consciousness assessment in rehabilitation training: An important aspect of autism treatment is rehabilitation training, such as social skills training, sensory integration training, etc. We plan to embed the artificial consciousness assessment module into the rehabilitation scene to monitor and evaluate the changes in children's cognitive status in real time, so as to achieve personalized intervention. Specifically, a cognitive assessment assistance system is developed: cameras and wearable devices are used to record the child's performance in the training session (such as facial expressions, heart rate, attention span), and the AI model analyzes the data in real time to determine the child's current mood and concentration level, as well as the level of understanding of the training instructions. This information is fed back to the therapist instantly, for example, when the system prompts "the child may feel frustrated and has decreased concentration at the moment", and the therapist can adjust the strategy accordingly (adding encouragement or taking a break for a while). In addition, based on data over time, the system can plot training progress curves, such as the trend of social eye contact frequency with training, which provides a quantitative basis for evaluating efficacy. In order to verify this function, we will select a certain number of trained children in cooperative rehabilitation institutions, half of them will be randomly assigned to use AI-assisted, and the other half will be trained as usual, and the difference in training effect will be compared. For example, the use of AI-assisted groups is expected to have a greater increase in social skills scale scores and a possible shortened training period. Such results will demonstrate that artificial awareness techniques can not only assess the problem, but also improve the intervention itself, thus highlighting the comprehensive value of the project system.

(3) Individualized intervention plan and multidisciplinary consultation: The best intervention for autism often needs to be individualized and multidisciplinary. This module will use our platform to suggest individualized interventions. After diagnosis, the system generates a comprehensive report based on each child's unique cognitive and neurological characteristics (e.g., weak language comprehension, good motor coordination, presence of OXTR gene variants, etc.), listing priority areas of intervention and options, such as: "Social communication training (improves eye contact); Oxytocin nasal spray was used to assist social motivation (due to low OXTR expression); sensory integration training (alleviating hyperacusis), etc." A multidisciplinary consultation (psychiatrist, rehabilitation therapist, genetic counselor, etc.) is then organized to develop a specific treatment plan based on the recommendations and discuss with the parents. In follow-up follow-up, the patient's progress data will continue to be entered into the system to update their understanding of the patient's status, and if an intervention is not effective, the system can also prompt adjustment of the plan. Through this closed loop, we explore the role of AI-assisted decision-making in personalized medicine for autism. The evaluation indicators can be parent questionnaire satisfaction, changes in children's core symptoms after intervention (such as improvement in social quotient assessment), etc. We expect that individualized approaches will achieve better outcomes and parental acceptance than traditional "one-size-fits-all" approaches. This will open up a new horizon for AI in decision support for autism treatment.

(4) Construction of autism artificial consciousness research platform: For long-term consideration, we will integrate the big data collected and the developed model algorithms throughout the project to build an open research platform. The platform includes: data layer (a database that stores brain imaging, genetic, behavioral multimodal data and follow-up information, and pays attention to patient privacy protection), model layer (an algorithm library that implements DIKWP models, which can be used to simulate different scenarios), and an interface layer (which provides a friendly web interface and API interface for researchers and clinicians to query and use). The platform initially serves the project team to manage the analysis process in a unified manner and accelerate the output of results. In the later stage of the project, we plan to gradually open up to a wider range of teams: invite domestic teams interested in autism research to join, share data resources and some models (with ethical permission), and regularly update the platform functions (such as adding analysis tools and visualization modules). At the same time, we will cooperate with relevant government departments or industry associations to strive to incorporate our platform into the industry infrastructure as part of the national autism big data center or intelligent diagnosis and treatment demonstration platform. This will greatly enhance the impact and vitality of the results of the project, so that it can continue to play a role after the end of the project, and promote new research and applications.

Expected Results: Driven by this module, we hope to achieve the "last mile" leap from laboratory results to clinical practice. Expected specific accomplishments include:

A set of empirical clinical diagnosis and treatment processes: autism assessment standard process integrating AI diagnosis, artificial consciousness-assisted rehabilitation training standards, multidisciplinary consultation and decision support guidelines, etc. These will be summarized as clinical guidance manuals, which may be submitted to the National Health Commission for trial on a larger scale, laying the groundwork for the development of industry standards.

Autism Artificial Intelligence Diagnosis and Treatment Platform (Software System): Software that integrates diagnosis, evaluation, intervention and recommendation functions, operates stably in pilot hospitals and institutions, and has scalability. By the end of the project, we plan to make it ready for rollout by reaching Technology Maturity Level 7 (demonstrated and validated in the relevant environment). Key modules of the platform, such as cognitive evaluation engine, AI diagnosis plug-in, data management system, etc., can be considered for registration as software copyrights or patents.

Clinical Data and Translational Evidence: The results of rigorously designed clinical studies demonstrate the effectiveness of our systems in improving diagnostic accuracy, reducing mean age at diagnosis, improving training efficacy, and high physician and patient satisfaction. This evidence will be published in the form of a paper and will be used to demonstrate the safety and efficacy of our technology to regulatory authorities, and to expedite approval and adoption.

Talent team and discipline development: During the implementation of the project, we have cultivated a group of compound talents who understand both artificial intelligence and brain science and clinical practice, including postdoctoral fellows and young doctors. This team will become the backbone of the research and service in the interdisciplinary field of "AI + autism" in the future, and also lay the foundation for related emerging disciplines (such as digital psychiatry and artificial intelligence medicine).

Overall, this module will ensure that the innovations of this project are truly rooted and translated into productivity for the benefit of patients and society. We not only provide "visible" new technologies, but also provide "well-used" new paradigms, so that artificial intelligence and artificial consciousness theory can more closely serve clinical needs and realize the transformation of science and technology into real productivity.

4. Feasibility analysis

The project is carried out by an interdisciplinary team led by Professor Duan Yucong, who has outstanding strengths in artificial intelligence, cognitive modeling, brain science and clinical medicine, which provides a solid foundation for the smooth implementation of the project. The following analyzes the feasibility of the project from the aspects of team foundation, technical conditions, and cooperation environment:

Comprehensive strength of the team: First of all, the project leader, Professor Duan Yucong, is the proposer of the DIKWP artificial consciousness theory, and has a world-leading academic position in the field of cognitive computing and artificial intelligence basic theory. As the first inventor, he has been granted 114 invention patents, covering many cutting-edge directions such as large model training, artificial consciousness construction, and cognitive operating systems. This means that the team has a large number of independent intellectual property rights and key technologies in the design of artificial intelligence algorithms and cognitive systems, and will not be controlled by others. In terms of brain science, the core members of the team include young researchers with a background in computational neuroscience and cognitive neuroimaging, who have presided over projects such as brain network analysis and brain-computer interface, and are familiar with the research methods of brain diseases. At the same time, there are also front-line clinical child psychiatrists and rehabilitation treatment experts, who have been engaged in the diagnosis and treatment of autism for a long time and have rich case resources and experience. Such a closely integrated lineup of "industry, academia, research and medicine" ensures the seamless connection of professional knowledge in all aspects of the project. For example, AI experts and clinicians can work together to develop feature sets when designing AI diagnostic models, and neuroscientists and AI experts can collaborate to incorporate biological patterns into the model when interpreting brain imaging results. The team also has rich experience in cooperation in the past, and has established a good communication and project management mechanism, with clear responsibilities and efficient cooperation for each sub-project. Therefore, from the perspective of staffing and organizational management, we are ideally positioned to take on this complex and cross-cutting topic.

Preliminary Research Basis: The team has carried out important preliminary work in related fields, providing preliminary results and technical reserves to verify the feasibility of the project:

In terms of AI cognitive models, we have developed a prototype system of DIKWP models, which deconstructs the reasoning process of large language models into five links: data, information, knowledge, intelligence, and intent, and monitors them. This proves that the DIKWP model can be embedded in the actual AI system and output interpretable results. This technology can be directly migrated to the design of the AI diagnostic module used in this project.

In terms of autism recognition algorithms, the team members recently published a paper in an international journal on the fusion of multi-perspective behavioral features to identify autism, and the proposed model has achieved leading accuracy on public datasets (such as using graph neural networks to fuse facial expressions and eye movements to achieve an accuracy rate of more than 85%). This shows that the team has a deep grasp of the difficulties and key points of AI diagnosis of autism, and has the ability to develop more complex models.

In terms of brain imaging and genetic research, we have preliminary fMRI data of children with autism, and conducted some network analysis, and found that the phenomena of autism such as reduced default network connection and sensory network overconnection are consistent with the literature. At the same time, we participated in the whole exome sequencing study of autism in China and found several risk gene variants unique to the Chinese population. These preliminary data support the hypothesis of the project study (e.g., autism does have abnormal brain connectivity and genetic basis) on the one hand, and on the other hand, provide data and experience for follow-up research, reducing uncertainty.

In addition, we have established cooperative relationships with well-known autism research institutions in China (such as the Autism Research Center of Peking University Sixth Hospital) and rehabilitation institutions, so that more samples can be obtained and joint research can be carried out. These collaboration channels ensure the source of cases and data required for the implementation of the project, and facilitate the replication of results.

Technical conditions and platform: The experimental conditions required for this project have been basically met in the unit where the team is located and the cooperative unit.

Laboratory equipment: Hainan University and cooperative hospitals have 3T magnetic resonance imaging instrument, 64-lead and above high-density EEG system, biological signal recorder, gene sequencing platform, high-performance computing server and other key equipment, which can meet the needs of brain imaging collection, genetic analysis and AI model training. Among them, the GPU cluster configured by the School of Computer Science (with a total computing power of hundreds of TeraFLOPS per second) will be dedicated to the training and simulation of large-scale deep learning models for this project.

Data resources: In addition to self-collection, we can also use authorized multi-center autism databases, including ABIDE brain imaging database, SFARI gene bank and other international resources, as well as behavioral video data accumulated by multiple rehabilitation institutions in China. These valuable data assets provide a broad foundation for model training and validation. We will also continue to gather more data in the construction of the platform to form a virtuous circle.

Core algorithms and tools: The team has mastered a number of self-developed software tools, such as cognitive graph construction software, brain network analysis package, federated learning framework, etc., which can be directly used in the development of this project. In addition, the team is proficient in using common AI and brain science tools such as TensorFlow, PyTorch, and SPM/fNIRS to speed up development. What's even more rare is that we have a code implementation and use license for DIKWP-related patents, which means that we can use these cutting-edge algorithm components in our projects without any barriers, and it is difficult for other teams to replicate this capability in a short period of time.

Clinical trial conditions: The partner hospital has passed the ethics approval, allowing us to introduce AI-assisted diagnosis and collect relevant data in the clinical environment. The hospital also said that it will open up its information system interface to facilitate our access to data, which solves the practical obstacles to the implementation of the AI system. In addition, in terms of gene editing animal experiments, the animal center of Hainan University is equipped with SPF-level animal rooms, microinjection, behavioral tests and other conditions, and our team members also have rich experience in mouse operation to ensure that the experiments are carried out safely and compliantly.

Project Risks and Countermeasures: Although the project is ambitious, we have fully considered the potential technical and implementation risks, and formulated a plan:

Data quality and sample size risk: Autism is heterogeneous and has high individual differences, which may make it difficult to train and generalize models. In this regard, we augmented the samples through multi-center and multi-modal data collection, and used data augmentation and domain adaptation techniques to improve the robustness of the model. At the same time, statistical experts are invited to control the experimental design to ensure that there is sufficient statistical power to detect the expected effect.

Interdisciplinary communication risks: AI engineers and clinicians have different understandings of problems, which can lead to communication barriers. In this regard, we regularly organize cross-cutting seminars and personnel exchange visits to allow members from different backgrounds to train each other on basic knowledge, establish unified terminology, and set up cross-topic leaders in project management to coordinate communication and ensure that the requirements are aligned.

Gene editing ethics and safety risks: Gene interventions are clinically strictly limited, we only explore at the cell and animal stages, all experiments adhere to ethical norms, and biosafety assessments are conducted in advance. We also pay attention to the international discussion of gene therapy for autism, and actively communicate with ethics experts to ensure that the study is within the acceptable range and is ready for possible clinical trials in the future.

Risks of technical implementation difficulties: for example, metacognitive artificial awareness modules may be difficult to debug, and the clinical adaptability of AI systems. In this regard, we have established milestones (see the next section) to ensure early verification of key technologies, and adjust the plan in time if the technical route is found to be unsuccessful. For example, if the metacognitive module is not effective, we can simplify it to the purpose verification of heuristic rules, which reduces the difficulty but still achieves some functions.

Based on the above analysis, it can be seen that the project is fully prepared in terms of personnel, finance, and materials, and there are countermeasures for key risks. We are confident that we will achieve the research objectives within the established timeline and flexibly adjust as needed in the process to ensure the success of the project and the high-quality results.

5. Milestones and milestones

To ensure that the project stays on schedule and produces high-level outcomes, we have developed a phased research plan with corresponding milestones. Each phase has clear mission objectives and expected outcomes for the project team and funders to track and evaluate. The following is a chronological list of the main milestones and milestones of the project:

**Phase 1 (initial stage of the project, about 1~1.5 years): theoretical framework construction and preliminary model validation. ***Milestone 1: DIKWP-Autism Cognitive Atlas Prototype Completed. *In this stage, we will complete the preliminary modeling of the multi-scale mechanism of autism, construct the prototype of the cognitive map of autism under the guidance of DIKWP, and write a theoretical paper. 【Outcome indicators】: A bilayer map model containing major brain regions, cognitive functions and potential gene nodes was formed; Validation of the atlas on a small sample can distinguish autism from normal, corresponding to the publication of a high-level review or model paper. At the same time, milestone *2: Prototype and interpretability verification of AI diagnostic model. *A prototype of a multimodal AI diagnostic model is preliminarily developed, the existing dataset is used for training verification, and an exemplary diagnostic report is output. [Achievement indicators]: The diagnostic accuracy of the prototype model exceeded 80%, and the interpretation report generated was recognized by more than 3 clinical experts; Apply for a model-related invention patent.

**Phase 2 (mid-project, about 2~3 years): core module R&D and pilot verification. ***Milestone 3: Autism Artificial Consciousness Diagnostic System Beta Built. *In this phase, we will integrate the multimodal data pipeline and the DIKWP artificial awareness architecture to develop a beta diagnostic system and conduct closed testing. [Achievement indicators]: The system integration has been improved, the performance has been verified on the data of no less than 100 local subjects, the diagnostic accuracy rate is more than 85%, and the interpretable report meets the clinical needs and improvement suggestions. Once this milestone is completed, deployment in pilot hospitals can begin. At the same time, *Milestone 4: Gene Function Validation and Animal Model Establishment. *The cell function experiment verification of key candidate genes was completed, and at least one gene was selected for animal experiments and the corresponding autism model mouse population was successfully constructed. 【Outcome indicators】: Cell experiments showed that editing the target gene could significantly change neuronal firing/synaptic behavior (P<0.05 significant), and the model mice showed the expected autistic behavior deficits; The results of the stage are published in the form of papers (e.g., the results of cell experiments are submitted in Molecular Psychiatry). In addition, *Milestone 5: The Optimized Version of the DIKWP Model Phase was released. *The data and discovery feedback obtained in the stage were optimized into the DIKWP cognitive model, and the DIKWP model v2 for the field of autism was launched. 【Outcome indicators】: The model added biological circuit parameters, which could simulate the occurrence mechanism of at least one core defect of autism; Write model improvement reports for internal publication and prepare them for public publication.

**Phase 3 (late stage of the project, about 4~5 years): integration platform improvement and clinical application verification. ***Milestone 6: The autism artificial awareness diagnosis and treatment platform was completed and put into trial operation. *Integrate the results of all modules, build a comprehensive platform combining online and offline, and start trial operation in cooperative hospitals/institutions. [Achievement indicators]: The platform has functions such as patient information management, multimodal data analysis, diagnosis reports, and biomarker recommendations, and has been deployed and trialed in no less than 3 institutions, serving more than 100 children in total, collecting feedback and iterating versions. *Milestone 7: Clinical Trial Completion and Data Analysis Report. *Complete the platform-based prospective clinical trial and intervention follow-up, systematically analyze all data, and obtain the clinical benefit evaluation of the project technology. [Outcome indicators]: Submit detailed clinical trial reports, including statistics on the improvement of diagnostic accuracy, the increase in the number of early screening cases, and the improvement of intervention effect, etc., to prove the superiority of the technology of this project compared with traditional methods (such as increasing the accuracy rate by X%, increasing the social quotient after the intervention, etc.); Published at least 2 key results papers in top journals (e.g., AI diagnostic system papers, gene therapy animal test papers). *Milestone 8: Development of standards and promotion plans. *At the end of the project, we will collate all the results and lessons learned, and prepare a draft technical standard and a promotion plan. [Outcome indicators]: Form the draft of the "Technical Specification for the Artificial Consciousness-assisted Autism Assessment System" and the "Guidelines for the Construction of Autism Artificial Intelligence Diagnosis and Treatment Center" and other documents, and communicate with relevant departments about the feasibility of incorporating them into the industry standard.

The above-mentioned milestones are interlocking and progressing gradually. Among them, milestones 1-3 are to lay the foundation for the follow-up of the project, milestones 4-5 mark that the project has entered the stage of tackling key problems and produce important scientific research results, and milestones 6-8 focus on transformation and application and closing summary. The setting of these milestones not only ensures the timely breakthrough of key technologies (such as AI systems and gene validation), but also ensures the implementation of application verification (such as clinical trials and platform operation), so as to achieve the goal of paying equal attention to scientific exploration and practical value creation.

Through the checks on the phased results, we are confident that the set goals of the project will be fully achieved. The results of each stage will also be used as a quantitative assessment point for the progress of the project, so that the fund management department can keep abreast of the project progress and give guidance to ensure the high-quality completion of the project.

6. Achievement form and assessment indicators

The expected outputs of this project will take a variety of forms, covering multiple levels such as theoretical methods, technical platforms and application demonstrations. In order to scientifically evaluate the effectiveness of the project, we have added a number of innovative indicators on the basis of the traditional assessment indicators and the characteristics of the project. The following is a summary of the main achievements of the project and the corresponding assessment indicators:

1. Theoretical and Model Results:

DIKWP Cognitive Model Extension and Theoretical Papers: The project will generate new extensions to the DIKWP model for autism, such as an artificial conscious cognitive framework that integrates brain connections and genetic factors. Assessment indicators: Publish X high-level academic papers (including at least 1 IF>10 review or theoretical paper) to explain the principles of the model and new explanations for the mechanism of autism; The model has been reported and exchanged more than Y times in international academic conferences, and has been positively evaluated by peer experts.

Cognitive Atlas and Knowledge Base: Construct a multi-level cognitive map and knowledge base for autism as an intermediate research achievement. Assessment indicators: 1 set of data/knowledge graphs with a complete structure , and the number of nodes and relationships reaches a certain scale (such as 100 brain nodes, 50 cognitive function nodes, and ≥≥50 gene nodes) ≥ brain nodes; The knowledge base accurately reflects the main signs and mechanisms of autism, and has been reviewed and approved by experts.

2. Technical Methods and Patented Software:

Artificial Consciousness Diagnosis Subsystem: The core algorithm module of AI diagnosis based on DIKWP. Assessment indicators: 1 set of DIKWP cognitive engine subsystem was developed, and its diagnostic accuracy was ≥ 90% (or at least 5 percentage points higher than that of the unexplained black box model), and the interpretation report was in line with clinical logic (expert satisfaction score ≥ 80%). In addition, the interpretive performance can be quantified, and if each case report contains an average of ≥3 verifiable evidence, the doctor's trust in the report score is ≥ 4/5.

Multimodal Data Fusion and Analysis Tools: Form reusable data analysis pipelines and software tools. Assessment indicators: at least 2 software copyrights or invention patents, such as "DIKWP-based multimodal autism recognition software V1.0", "autism cognitive graph construction method" patent, etc. The software is installed and run in the cooperative unit, and the user feedback is good.

Gene Editing Experimental Method Patent: Editing intervention method or product designed for the screened autism target gene. Assessment indicators: 1 invention patent ≥, such as "XXXX gene splicing therapy method for improving social behavior of autism", and entered the substantive examination. Although it is difficult to be clinically implemented in the short term, it is novel and feasible as a reserve technology.

3. Platform and Application Results:

Autism Artificial Awareness Research and Diagnosis and Treatment Platform: An integrated platform that includes a database, analysis system, and user interface. Assessment indicators: the platform has complete functional modules, and there are no major failures after pilot operation; The number of cases included in the platform ≥ N cases (e.g., 200 cases ≥), including multimodal data and follow-up information; The number of platform users (number of investigator or doctor accounts) ≥M. An important part of the platform, such as the cognitive assessment engine, is recognized through third-party assessments.

Demonstration application and clinical report: Realize the demonstration application of AI-assisted diagnosis in partner hospitals. Assessment indicators: 100 clinical diagnosis reports ≥ issued, and the accuracy and quality of the reports meet the requirements of doctors; Compared with before the start of the project, the average age of diagnosis in the pilot hospitals was X months earlier (or the missed diagnosis rate was reduced by Y%); With the assistance of rehabilitation institutions, ≥ Z copies of training evaluation reports, and parents' satisfaction was improved. One clinical research report**/white paper was formed to summarize the clinical performance and improvement direction of the AI system.**

Standards and Guidelines: The results are to be elevated to industry standards or expert consensus. Assessment indicators: Lead/participate in the formulation of 1 group standard or guideline ≥, such as "Technical Guidelines for Intelligent Diagnosis and Assessment of Autism in China", and submit it to the society/association for discussion. If it can be approved for release within the project period, it will be considered overfulfilled.

4. Talent Development and Team Building:

Cultivating a number of cross-border talents. Assessment indicators: ≥ x dissertations of doctoral/master's degree dissertations are completed on the basis of the results of this project; 5 project team members made presentations at international conferences≥ Some team members have been selected into the high-level talent plan or won relevant scientific and technological awards. Team cohesion and continuous research ability have been enhanced, and a stable interdisciplinary research direction has been formed.

5. Public Perception and Social Impact:

This project will raise public awareness of autism and artificial intelligence through popular science and media outreach. Assessment indicators: 5 ≥ popular science lectures**/trainings**; Mainstream media reported on the progress of the project ≥ 3 times; At the end of the project, the public (especially families with autism) had significantly increased their acceptance of AI-assisted diagnosis and treatment (before and after through a questionnaire).

Among the above indicators, there are quantitative indicators (number of papers, accuracy improvement, etc.) to ensure objective measurement, and qualitative indicators (expert satisfaction, standard setting, etc.) to reflect the long-term value of the project. Particular emphasis was placed on two new indicators:

Artificial consciousness subsystem construction indicators: Requires the successful development of the DIKWP artificial consciousness module for the diagnosis of autism, which will verify the existence in the form of patented software or technical reports. Its performance is quantified by explanatory and controllability indicators, such as the length of the decision-making traceability chain, the success rate of self-regulation of anomaly detection, etc., which will be refined during the development process.

Quantitative indicators of DIKWP cognitive engine for diagnosis improvement: Compared with the models with or without DIKWP framework, the differences in diagnostic accuracy, false positive and false negative rate, and model correction ability were measured. For example, we will report that "after the introduction of DIKWP, the model increased the recall rate for mild cases by X% and reduced the misjudgments for confounding diseases such as ADHD by Y%". Another quantification is the correctness of interpretation: the proportion of key conclusions in AI reports that are consistent with those of physicians, which is expected to be significantly higher than that of models without DIKWP.

Through these indicator systems, we can comprehensively evaluate the scientific and technological output and application value of the project. We will regularly check our progress against indicators, identify deviations and adjust our priorities to ensure that the final deliverables are of high quality and stand up to scrutiny. For example, if the output of the paper does not meet expectations at a certain stage, we will strengthen the data analysis and writing organization; If the diagnostic accuracy is not significantly improved, the model design can be troubled or the amount of data will be increased. The assessment indicators are both a constraint and a motivation, prompting the team to maintain high standards, ultimately achieve the expected goals of the project, and even achieve gains that exceed expectations.

7. Campaigns and conversion paths

The results of this project have broad application prospects and industrial transformation value. We have formulated a clear promotion plan and path to ensure that after the completion of the project, these new technologies and paradigms can quickly move from scientific research to clinical and market, and produce tangible social and economic benefits.

1. Establish a research platform for artificial awareness of autism and promote continuous innovation: One of the results of the project is the research platform for artificial awareness of autism (see module 3.4 for details), which is the basis for future promotion. At the end of the project, we will set up a "Research Center for Artificial Intelligence and Consciousness for Autism" at Hainan University or a partner institution, and continue to operate on this platform. The center will integrate research, clinical, and industry, and continuously update data and algorithms, so that the platform will become a resource shared by autism researchers and doctors across the country. We plan to apply for the platform to be included in the management of the national or provincial science and technology infrastructure platform to obtain policy and financial support. In the long term, the platform can also be expanded to include modules on other developmental disabilities (e.g., ADHD, language delays) to become a comprehensive AI research infrastructure for child developmental disabilities. Through the openness and scalability of the platform, it attracts more teams to join, continuously incubates new research topics and technology applications, and forms a virtuous circle of new qualitative productivity platform.

2. Model as a service deployment to accelerate clinical popularization: After successful clinical validation, we will package the AI diagnostic model and cognitive assessment system and deploy it to the cloud to provide software-as-a-service (). SaaS). Hospitals or rehabilitation facilities do not need local complex equipment, and can simply upload patient data through a secure interface to obtain AI analysis results. This greatly lowers the threshold for the use of technology, which is conducive to the promotion of medical and rehabilitation units at all levels across the country. We will work with healthcare IT companies to integrate the service into their existing electronic medical records or diagnosis and treatment systems and into the physician workflow. For example, we worked with a major children's hospital information system provider to embed our AI assessment into the electronic record interface of the autism clinic so that doctors can call up the analysis with one click. For grassroots or remote areas, we also plan to develop mobile apps or portable terminals so that grassroots doctors or parents can also access initial screening and guidance. Through the deployment of cloud services and apps, we expect to achieve autism diagnosis and treatment centers in major provinces across the country within 3 years after the end of the project, so that more than 1,000 doctors can use our system, and more than 10,000 children will benefit from it.

3. Industry standardization and certification promotion: We are well aware that medical AI products must obtain industry standards and regulatory certifications for widespread application. To this end, we will actively participate in the development of relevant standards (such as the draft guidelines/standards mentioned in the aforementioned indicators). When the project is completed, we strive to make "AI-assisted autism diagnosis" written into national or industry standards. In terms of supervision, we will apply to the State Food and Drug Administration for approval of AI medical devices on the basis of clinical trial data, and strive to become the first approved AI diagnostic software for autism in China. At the same time, we will promote the inclusion of our assessment indicators in the diagnosis and treatment of autism, such as suggesting that AI objective assessment be used as an auxiliary diagnosis basis. These efforts will lay the foundation for market access and regulated use. Once we have obtained certification and standard endorsement, our technology will be more easily procured and adopted by medical institutions around the world, accelerating the speed of adoption.

4. Commercialization and industrial implementation: Although this project is led by scientific research, we attach great importance to the industrialization potential of the results. The patent pool of Professor Duan Yucong's team has built an intellectual property moat in the field of AI cognition and artificial awareness. Relying on this, we will explore the establishment of technology start-ups or joint existing enterprises for industrial transformation. Specific paths include: negotiating cooperation or technology licensing with large medical device/medical AI companies, and expanding AI diagnostic systems as their product lines; Cooperate with genetic testing companies to develop a package of solutions for autism, including genetic testing + AI assessment + consulting services; Seeking venture capital investment, the company established a start-up company to focus on the development of children's intelligent diagnosis and treatment platform. If the conditions are ripe, the core members of our team can start a business part-time to ensure that the core technology does not leak out. With the abundant patents that Professor Duan has obtained, we have a strong advantage in the negotiation to attract corporate investment and move forward together. Once industrialized, the company's profit model can be software licensing, subscription services, supporting hardware sales, etc. It is conservatively estimated that there are millions of children with autism and related disorders in China, and no less than 100,000 new diagnostic needs are added every year, and the market demand for intelligent diagnosis and treatment systems is huge. On a per-case basis, the potential market size is billions of dollars per year. At the same time, the results of genetic interventions also have patent value, which can be licensed to pharmaceutical or biotechnology companies to promote drug or gene therapy research and development. Overall, the commercial prospects of the project results are broad, and we will carefully plan the patent and commercial layout to truly transform the scientific research results into productivity and emerging industries.

5. Social impact and policy support: To drive transformation, we will also start with the social and policy aspects to create a positive environment. On the one hand, we will convey the value of our technology to decision-makers through industry conferences and media reports, and strive to incorporate "artificial intelligence + autism" into relevant policies (such as digital health projects, rehabilitation support programs for disabled children, etc.). On the other hand, we should actively carry out science popularization to improve the public's acceptance of AI diagnosis and treatment and reduce possible misunderstandings and resistance. We will also maintain high standards in ethics and safety to earn trust with practical results. Through close contact with the government and associations, we hope to promote demonstration projects, such as the pilot of "AI early screening for autism" in a province, where the government will purchase services to provide screening for high-risk children. This can not only quickly expand the impact, but also reflect the public good value of technology.

6. Expand to related fields: The artificial consciousness cognitive framework and intelligent diagnosis and treatment platform built by the project results are not limited to autism as a disorder. After success, we plan to replicate the experience to other neuropsychiatric diseases, such as ADHD in children, depression, Alzheimer's disease, etc., all of which have the problem of relying on subjective diagnosis and lacking objective indicators, and can also introduce the DIKWP model for improvement. We are already exploring collaboration with the Alzheimer's Diagnosis and Treatment Center to use artificial awareness assessment for mild cognitive impairment screening, for example. If this expansion goes well, our technology platform and company product line will be richer, so as to obtain greater social and economic benefits. This horizontal outreach will also feed back into the autism community, further enhancing the maturity and reputation of our approach.

To sum up, the promotion and transformation plan of this project is clear and comprehensive, emphasizing the four-pronged approach of platform construction, service deployment, standard guidance, and commercial operation. At the same time as the completion of the scientific research task, we have paved the way for its subsequent development, and strive to achieve "research and application". It is foreseeable that at a time when the relevant policies and market environment are increasingly supporting the "transformation of scientific and technological innovation", the results of this project will be able to be implemented quickly and achieve large-scale application within five years。 It will significantly improve the scientific and technological level of autism diagnosis and treatment in China and produce considerable social benefits. At the same time, it will give birth to new industrial growth points, form internationally competitive intelligent medical products and services, and add innovation momentum to China's artificial intelligence and biomedical industry. This is in line with the country's requirements for building a "new quality productivity platform" - through original innovation, create a new type of scientific and technological productivity with cross-border integration, and promote high-quality development. The successful roll-out of this project will be a vivid example of this concept.

In summary, we are confident in the research plan, implementation capability and expected outcomes of this project. With the support of major national scientific and technological tasks, the research on the neural mechanism of autism and gene editing therapy integrating the DIKWP model will surely produce a number of breakthrough results, create a new situation in the research of autism and related brain diseases, and transform scientific discoveries into new means and products that benefit people's health with solid transformation measures. We kindly ask all experts to review the application of this project and give valuable advice and support!

References:

· Fernández M, et al. Neural Circuits for Social Cognition: Implications for Autism. Neuroscience. 2018. (A review of the role of social brain networks in autism)

·· Hajri M, et al. Cognitive deficits in children with ASD: integrative approach. Front Psychiatry. 2022. (A review of social and non-social cognitive impairments in autism)

·· He C, et al. Identifying ASD from multi-modal data with hypergraph neural networks. NPJ Mental Health Res. 2024. (Multimodal Fusion AI Method for Diagnosing Autism)

·· Phoenix.com. Prof. Duan Yucong: DIKWP Artificial Consciousness Model Leads the Future of AI. 2025. (DIKWP Model Theory and Patent Achievement Report)

·· The Transmitter. CRISPR therapy may reverse autism mutation’s effects. 2019. (CRISPR reversal of autism-related gene mutation in mouse phenotype)

·· Srivastava A, et al. Therapeutic strategies for autism: targeting central dogma. Transl Psychiatry. 2023. (Review of Gene Therapy Strategies for Autism)