Project background and existing technology gap analysis

Movement disorders such as Parkinson's disease (PD) are common degenerative diseases of the central nervous system in middle-aged and older adults, characterized by progressive degeneration of dopaminergic neurons, resulting in symptoms such as tremor, bradykinesia, muscle rigidity, and abnormal postural gait. At present, the prevalence of these diseases is increasing year by year around the world, seriously affecting the quality of life of patients and placing a heavy burden on society and families. Although treatments such as drugs (eg, levodopa) and deep brain stimulation (DBS) can partially alleviate symptoms, they do not stop disease progression, and long-term drug use can fluctuate and become resistant, and DBS surgery is costly and limited. The existing intervention methods are mainly to passively respond to the symptoms of the disease and treat them after the symptoms appear, and there is a lack of active prevention and intervention mechanisms for the early stage of the disease or even before the occurrence of the disease. This "passive medicine" model has obvious drawbacks: patients are often diagnosed when neurons are dying massively and their function is severely impaired, missing the best time to intervene. At the same time, the complex interactions between various pathogenic mechanisms have not yet been fully understood due to the complex etiology of movement disorders, involving multiple factors such as genetic factors, environmental toxins, aging processes, and imbalances in the metabolism of neurotransmitters such as dopamine. Traditional biomedical research and clinical trials often focus on a single level of mechanism, and it is difficult to fully reveal the whole process of the disease. In addition, the application of existing AI technology in medical treatment mostly stays at pattern recognition and auxiliary diagnosis, and there is still a lack of systematic methods for modeling disease mechanisms, simulating the evolution of disease course, and guiding individualized treatment. This technological gap makes it urgent to explore new interdisciplinary solutions, integrate the cutting-edge advances of neuroscience and artificial intelligence, and establish an intelligent intervention system under the paradigm of active medicine.

The concept of "proactive medicine" advocates moving the medical threshold forward, and through continuous monitoring and proactive intervention, action can be taken before the disease occurs or at an early stage to prevent minor illnesses from becoming major diseases. This concept requires the integration of multi-source data, knowledge and artificial intelligence technology to achieve global control and dynamic management of personal health status. In the management of degenerative diseases, active medicine emphasizes the combination of early detection, individualized treatment and long-term management, and empowers doctors and patients to make collaborative decisions through AI, moving the "treatment window" forward. However, active medical research on movement disorders such as Parkinson's disease is still in its infancy: we lack effective models to comprehensively analyze data from patients at all levels, from molecular and neural circuits to clinical symptoms, and there is a lack of platforms that can reconstruct disease mechanisms and test treatment options in a virtual environment. In the field of neuroscience, there are a large number of research results on motor control and Parkinson's disease mechanisms (such as models on [the function of basal ganglia circuits]{.underline}), but these knowledge exist scattered in the literature, and a unified semantic model has not yet been formed to facilitate computer understanding and reasoning. Although deep learning, reinforcement learning and other methods have emerged in the field of artificial intelligence, traditional AI is mostly a black-box model, lacking the explainability of the decision-making process and the rigorous reasoning required in the medical field. At the same time, a purely data-driven approach is difficult to directly apply to small-sample, high-complexity medical problems. Recent studies and reviews have pointed out that computational modeling tools are essential to navigate the complex pathological network of movement disorders such as Parkinson's disease, and multi-scale models across all levels of pathogenesis, neurodynamics, and treatment strategies are needed to guide us beyond the limitations of intuition. For example, some scholars have used computational models to reveal the origin of abnormal oscillations in the β band (15–30 Hz) in the basal ganglia network of Parkinson's disease, and some studies have used simulations to better understand the mechanism of action of DBS treatment. These all show that system modeling is of great value in unraveling the complexity of movement disorders. However, at present, such models are mostly limited to specific subsystems or single scales (such as only neural circuit models or molecular network models), and have not yet formed a unified framework for integrating cognitive semantics and physiological mechanisms.

In order to bridge the above technical gaps, we plan to introduce a series of original theories and technologies proposed by Professor Yucong Duan, including the DIKWP artificial consciousness model, ACPU architecture, [semantic elastic network]{.underline} and self-explanatory mechanism, etc., to integrate neuroscience and artificial intelligence to build an intelligent and active intervention medical platform. The DIKWP model is an extension of the classic DIKW (pyramid) model, adding the fifth layer of "Purpose" on top of the four layers of Data, Information, Knowledge, and Wisdom, and replacing the linear level with a network structure, so that the semantics of each layer can be fed back and updated iteratively in both directions. This innovative cognitive system provides a full-link traceable semantic representation for the AI decision-making process, which is expected to solve the problem that traditional black-box models are difficult to interpret and control. As Professor Yucong Duan explains, "The DIKWP model builds a common cognitive language between humans and machines, allowing every step of the AI decision-making process to be traced, explained, and understood by humans." By embedding the key layer of 'purpose' inside the model, we are not only able to make AI smarter, but also ensure that it always serves human values and security needs." The semantic elastic network based on DIKWP can organize massive heterogeneous knowledge in the medical field in the form of multimodal semantic graphs, and realize the flexible association of different levels of information through elastic links, so that the model can dynamically adapt to the integration of new knowledge and environmental changes. This opens up the possibility of real-time knowledge updates and personalized interpretations required for active medicine. [The Artificial Consciousness Processing Unit]{.underline} (ACPU) is a brain-inspired computing architecture that is proposed in conjunction with it It aims to implement the DIKWP model at the software and hardware levels, and support real-time cognitive loop and metacognitive feedback. In short, ACPU is equivalent to the brain processing unit of artificial intelligence, and its architecture design integrates subliminal computing (such as pattern recognition ability of large model-like LLM) and conscious DIKWP semantic reasoning mechanism, so as to improve the autonomous decision-making ability and efficiency of artificial intelligence in complex and dynamic environments. Simulation results show that the real-time bidirectional fusion mechanism of deep semantic space and high abstract concept space can be significantly improved by designing a real-time two-way fusion mechanism of deep semantic space and high abstract concept space in ACPU, and the quality of Wisdom decision-making can be significantly improved, and human-like cognitive flexibility can be given to artificial intelligence systems. Finally, the self-explanatory mechanism runs through the above models and architectures to ensure that the system can give an intelligible explanation based on the DIKWP semantic link when making decisions or intervention suggestions. For example, for the treatment plan recommendation for patients with Parkinson's disease, the system can explain the basis: starting from the patient's data (symptoms, examination indicators), extracting key information and patterns (information layer), combining medical knowledge and previous cases (knowledge layer) to analyze possible pathological mechanisms, and weighing factors such as long-term prognosis and quality of life ( Wisdom layer), which ultimately gives recommendations for intervention and its intended purpose (Purpose layer), and is able to give human-understandable justification and justification for each step of reasoning. This self-explanatory capability is particularly important for medical AI, as it not only increases the trust of doctors and patients in AI decision-making, but also facilitates expert review and correction of system recommendations.

In summary, this project proposes a plan for the construction of an active medical research and intervention system integrating the DIKWP model in view of the key shortcomings in the current research and intervention of movement disorder neurological diseases. Through systematic analysis of the existing technology gaps, it can be seen that a platform that can integrate multi-scale neuroscience mechanisms and interpretable AI semantic models is needed to reconstruct the pathogenic process of Parkinson's disease and other diseases, and realize the virtual testing and optimization of intervention methods such as drugs and devices, and at the same time support brain-computer interfaces to realize the reconstruction and feedback control of patients' motor purpose. In the following sections, we will elaborate on the six core research contents of the project: 1) theoretical modeling of motion control and neurodegenerative processes based on DIKWP semantic system mapping; 2) Design of ACPU-driven brain-like circuit regulation simulation platform; 3) Parkinson's disease and other mechanism reconstruction and drug testing environment supported by the virtual patient system; 4) brain-computer interface and motor purpose reconstruction mechanism; 5) Domestic active medical intervention program, drug screening and intervention device design; 6) Phased tasks and key outcome nodes of the three-to-five-year project. Through the integration of these innovative points, we hope to design a set of intelligent active intervention medical platform architecture with an engineerable path (can be implemented), an evaluation system (measurable effect) and sustainable evolution ability (iterative optimization), and create a new paradigm for the prevention and treatment of movement disorder neurological diseases.

The DIKWP semantic system models the mapping of motor control and neurodegenerative processes

Motor control is a multi-layered and complex process, involving the production of higher levels of purpose in the cerebral cortex, to the selection and initiation of motor patterns by central structures such as the basal ganglia, to the spinal cord and peripheral nerves to perform specific muscle actions, and sensory feedback from proprioception and vestibular and visual feedback. In movement disorders such as Parkinson's disease, various parts of this process are affected to varying degrees, eventually leading to motor dysfunction. We plan to construct a DIKWP semantic system to comprehensively map the semantic hierarchy and mechanism of this process, and organically combine the biomedical mechanism with cognitive semantics to form a unified representation of motor control and neurodegeneration.

Firstly, the hierarchical meaning of the DIKWP model is briefly described: the data (D) layer corresponds to the original biological signals and objective data, such as electromyography signals, EEG/EEG signals, brain structure images, biochemical indicators, etc., which constitute the underlying record of the state of the motor system. In healthy people, the data layer covers a large number of signals generated by various links during the execution of the exercise plan (such as the original cortical firing pattern, the frequency of basal ganglia firing, the strength of muscle contraction, etc.). The information (I) layer is the meaningful patterns and features extracted from the processing of the data layer, such as the characteristic parameters of motion-related potentials, the intensity of neural oscillations in specific frequency bands, and the motor performance indicators (gait speed, step length, etc.). Through the information layer, we transform the messy data into indicators and signal characteristics that can reflect the state of motor function. For example, the enhancement of synchronous oscillations in the β band (~20 Hz) in the basal ganglia loop is a typical information-layer feature of Parkinson's disease, suggesting enhanced motor inhibition and impaired motor initiation. The knowledge (K) layer further combines the domain knowledge of medicine and biomechanics to elevate the characteristics of the information layer to the understanding of the mechanisms of fighting disease and the principles of motor control. This includes knowledge of the function of the anatomical circuit (eg, we know that the direct striatum-pallidus pathway activation promotes movement, and the indirect pathway activation inhibits movement; Dopamine has an excitatory effect on the direct pathway and an inhibitory effect on the indirect pathway, etc.), including the existing knowledge of the pathogenesis of Parkinson's disease (e.g., abnormal aggregation of α-synuclein leads to the death of dopamine neurons in the substantia nigra, which in turn causes striatal dopamine insufficiency). The knowledge layer allows us to explain the reasons behind the anomalous features that appear in the information layer. For example, we know that enhanced oscillations in the basal ganglia β are associated with dopamine deficiency and overexcitability of the indirect pathway, so that we understand the pathological mechanism behind this informational feature and guide interventions accordingly (e.g., the use of electrical stimulation to inhibit the overactivity of the indirect pathway node STR). The Wisdom (W) layer introduces high-level considerations such as clinical decision-making experience, ethics, and long-term effects. At the Wisdom level, we evaluate and make decisions on the possible interventions provided by the knowledge layer, taking into account factors such as the overall condition of the patient, the risk-benefit of treatment, the humanistic care, and the quality of life. For example, for a patient with Parkinson's disease, the knowledge layer may provide several intervention ideas: increasing the dose of drugs, implementing surgical intervention (DBS), or trying exercise rehabilitation. The Wisdom layer will take into account the patient's age, disease course, financial status, personal preferences (such as whether they are willing to undergo craniotomy), as well as long-term efficacy and side effects, etc., to make an individualized and best decision plan that is in line with the overall interests of the patient. The Purpose/Purpose (P) layer defines the ultimate goal, i.e., the goal of improving motor function and quality of life for the patient. The purpose layer includes both short-term goals (eg, elimination of tremors, improved gait) and long-term visions (eg, slowing disease progression, maintaining the ability to live independently). This level ensures that the entire package of interventions is centred around the stated medical objectives and provides a yardstick by which to evaluate the effectiveness of the intervention.

Based on the above definitions of DIKWP, we map the motor control process and the degenerative evolution of Parkinson's disease to the semantic space of DIKWP. Specifically, in healthy motor control, the layers are closely connected: the Purpose layer is generated by the brain (e.g., the purpose of "reaching out for a cup"), the Wisdom layer evaluates the necessity and safety of this action (e.g., ensuring that other objects are not knocked over, conforming to social situations, etc.), the knowledge layer invokes the previously learned motor skills and internal models (how to coordinate the arm muscles to complete the action of holding the cup), and the information layer is transformed from motor instructions into neural firing patterns and muscle recruitment patterns. The data layer is the occurrence of specific physiological electrical signals and muscle movements. When the exercise is successfully completed, the information of each layer is also corrected through sensory feedback, forming a closed-loop regulation.

In the case of movement disorders such as Parkinson's disease, we can use the DIKWP model to describe the dysregulation of each layer: first, in the data layer, the loss of dopamine neurons leads to a significant decrease in the intensity of the dopamine signal received by the striatum, which is a change in the original physiological data; At the same time, abnormal neural synchronization (e.g., decreased theta waves, hypersynchronization of β waves) can occur in EEG/local field potential recordings, and EMG data show decreased muscle activity at the initiation of movement, etc. These data-level abnormalities document objective physiological changes in Parkinson's disease. At the information level, these data are reflected as specific abnormal patterns: for example, the continuously high amplitude of the basal ganglia β oscillation is extracted as an information feature, indicating enhanced motor inhibition; For another example, gait analysis information may reveal that patients have characteristics such as smaller stride length and procrastinating steps; The olfactory test information showed that the patient's sense of smell decreased significantly, etc. Each piece of information corresponds to a signal of the underlying pathological process. The knowledge layer provides an explanation for this: according to medical knowledge, we explain the enhancement of the oscillation of the basal ganglia β as a relatively strong indirect pathway and a dysregulated loop rhythm due to dopamine deficiency; Reduced gait and shuffling are understood to be clinical manifestations of akinesia, resulting from impaired midbrain pacemaker function; Olfactory decline is associated with prodromal pathological changes such as possible Lewy bodies in olfactory bulb accumulation by the knowledge layer. The knowledge layer also includes an integrated understanding of pathogenic mechanisms, such as the α-synuclein misfolding to form Lewy bodies in Parkinson's disease, the role of neuroinflammation, and how genetic susceptibility (mutations in genes such as LRRK2 and PINK1) affect autophagy and mitochondrial function. This knowledge helps us model the causal network of disease. In the management of Parkinson's disease, the wisdom layer is reflected in the high-level design of clinical strategies and rehabilitation goals, including when to intervene in drug therapy, when to consider surgery, how to balance the improvement of motor symptoms with drug side effects, and the consideration of psychological and social support for patients. At the Wisdom level, we emphasize patient-centered long-term planning, such as early motor function exercise programs to slow down muscle wasting and joint stiffness, with the aim of "delaying disability"; Consider the patient's lifestyle and compliance in the drug regimen, and develop a simplified and effective medication plan. The purpose layer clarifies the ultimate goal of the whole intervention, such as "allowing patients to maintain their ability to walk independently and avoid serious falls in the next 3 years", "maximizing the improvement of self-care and dignity", etc. These purposes guide the development of strategies at the Wisdom level and become the ultimate criteria for evaluating efficacy.

In order to use the DIKWP model to formally model movement disorders, we will use semantic networks and ontological methods to construct a knowledge graph related to Parkinson's disease. This is done by grouping concepts, entities and processes related to motor control and Parkinson's disease into the atlas according to the hierarchy of DIKWP. For example, at the data layer, we define various data nodes (e.g., "EMG signal", "dopamine concentration", "gait video"); At the information layer, define state**/feature nodes** (e.g., "basal ganglia β oscillates too strongly", "Bradykinesia", "hyposmiculture"); The knowledge layer includes mechanism**/diagnostic nodes** (e.g., "X% loss rate of dopamine neurons in the substantia nigra compacta", "inhibition of striatal D1 receptor signaling pathway", "middle Parkinson's disease (Hoehn-Yahr class III)", etc.); The Wisdom layer contains decision**/strategy nodes** (e.g., "Adjust levodopa dose", "Recommend DBS surgical assessment", "Intensive gait training"); The Purpose layer includes target nodes (e.g., "50% symptom reduction", "reduction in risk of falls within one year", "maintenance of self-care ability"). These nodes are connected by directed correlation edges, forming a causal and semantic network. For example, "loss of dopamine neurons in the substantia nigra (knowledge)" points to "decreased dopamine concentration (data)" through the causal side, which in turn explains "enhanced oscillation of the basal ganglia β β (information)", "enhanced oscillation of the basal ganglia" further leads to "slow movement (information)", whereby the doctor will diagnose "intermediate Parkinson's disease (knowledge)" at the knowledge level, and then the "brain pacemaker treatment (decision-making)" will be selected by the wisdom layer to achieve "improved motor function (purpose)", etc. Through such a semantic network, we can globally represent the chain of Parkinson's disease onset and intervention. It is worth emphasizing that this is not a static tree-like structure, but a mesh-like elastic semantic network: there is a two-way connection and cyclic feedback between the nodes of each layer, which is in line with the "dual loop" concept of the DIKWP model. For example, when we implement an intervention (e.g., DBS), the data layer will generate new signal changes, the information layer will refine features such as "reduced tremor amplitude", and the knowledge layer will update the assessment of the condition (symptom improvement) accordingly, and the wisdom layer will adjust the follow-up treatment strategy and even update the overall recovery goal (Purpose). In this way, a closed-loop adjustment is formed between the upper and lower layers: on the one hand**, the purpose (P) layer drives the actions of the lower layers, and on the other hand, the change of data to the Wisdom layer will in turn affect the purpose** and achieve continuous optimization. For example, an objective improvement in a patient's motor function (data/information layer) may motivate a patient to be more willing to recover (Purpose layer) and adjust to new health goals.

By mapping and modeling motor control and disease processes through the DIKWP semantic system, we will theoretically gain a global understanding of movement disorders such as Parkinson's disease: both signals and pathologies at the physiological level, as well as goals and goals at the cognitive decision-making level. On the one hand, this model can be used as the semantic basis for subsequent simulation and intervention systems, and on the other hand, it can also be independently used for inference analysis and early warning of disease mechanisms. For example, we can use this model to interpret the semantics of early Parkinson's disease: before the clinical symptoms appear, there may be abnormalities in the information and data layers, such as decreased sense of smell, abnormal sleep behavior (REM sleep behavior disorder), etc. The model is able to correlate this fragmented information to potential disease risks (e.g., α-synuclein pathology in progress) through the knowledge layer, and the Wisdom layer can then recommend early life interventions or pharmacotropic strategies (e.g., regular exercise, neurotrophic supplementation) to serve the purpose of "delaying the onset of Parkinson's disease". This embodies the spirit of proactive medicine: proactive intervention before the disease has a full-blown outbreak. For example, for patients who have already been diagnosed, the model can analyze the status of each layer: if a patient is identified as belonging to the "drug fluctuation stage" in the knowledge layer, the Wisdom layer can decide to introduce adjuvant therapy, such as DBS or COMT inhibitors, and "maintain symptom stability" as the new purpose goal. In conclusion, DIKWP semantic modeling provides a hierarchical analysis and synthesis platform for movement disorders, enabling us to dissect disease processes like "dissecting" computer programs, focusing on both the underlying data and the high-level purpose, laying a theoretical foundation for active intervention.

Design of ACPU-driven brain-inspired circuit regulation simulation platform

After establishing the DIKWP semantic model, a platform capable of realistically simulating brain circuitry and disease dynamics in a computer environment was needed to validate theories, discover new mechanisms, and test intervention strategies. In this project, we propose to design an ACPU-driven brain-like circuit regulation simulation platform using the Artificial Consciousness Processing Unit (ACPU). ) to model and modulate key neural circuits related to motor control. The platform will serve as a "virtual brain bench" in which researchers and clinicians can reconstruct neural circuit abnormalities in movement disorders such as Parkinson's disease, test the effects of various electrical stimulation, medications, and behavioral interventions on circuit dynamics and functional output, and accelerate the innovation process from mechanistic studies to therapy design.

First, explain the role of the ACPU architecture in this platform. ACPU is a new computing hardware and architecture concept proposed by Yucong Duan's team, which aims to simulate the mechanism of the conscious and subconscious working together in the human brain. To put it simply, ACPU consists of two main subsystems: one is the subliminal computing subsystem, which is good at massively parallel data processing and pattern recognition (similar to the unconscious information processing of the human brain, which can be implemented by deep learning networks or brain-like chips); The second is the consciousness computing subsystem, which is based on the DIKWP model for high-level semantic understanding, reasoning and decision-making (corresponding to the conscious thinking part of the human brain). These two subsystems are coupled together by a bi-loop architecture: the subconscious subsystem continuously converts the environment and internal data into information and provides it to the conscious subsystem; The consciousness subsystem selectively pays attention to this information according to the purpose drive, integrates knowledge, and makes wisdom decisions, which in turn guides the further information collection and processing of the subconscious layer. In terms of hardware, ACPU can be implemented as a hybrid architecture that fuses brain-like chips (e.g., synaptic devices, neural network-like acceleration chips) and symbol processing units, as well as corresponding software (Artificial Consciousness Operating System, ACOS). It is characterized by real-time (it can respond to changes in sensor data in real time), autonomy (built-in goal-driven cyclic feedback), and interpretability (using the DIKWP model to record the decision-making process). Using ACPU in our platform is like equipping the virtual brain with an artificial "frontal lobe", which can regulate and plan the simulated brain circuit activity like the human brain, so as to achieve a more intelligent simulation than traditional pure algorithm simulation.

At the heart of the platform is a multi-scale brain network simulator where we will focus on simulating neural circuits that are closely related to movement disorders. The most critical of these is the basal ganglia-thalamic-cortical circuit, which plays a decisive role in the initiation and regulation of voluntary movements and is the main target of Parkinson's disease lesions. The basal ganglia mainly include structures such as the striatum, the inner and outer globus pallidus (GPi/GPe), the substantia nigra dense and reticular parts (SNc/SNr), and the thalamic basal nucleus (STN), which form multiple closed-loop pathways (direct, indirect, and superdirect) with the cerebral cortex and thalamus. In healthy conditions, the cortex excites the striatum, and the striatum inhibits GPi/SNr through direct pathways, thereby relieving the inhibition of GPi/SNr on the thalamus and promoting movement; On the other hand, the striatum excites GPe through indirect pathways, thereby inhibiting the excitatory effect of STN on GPi/SNr, indirectly weakening the ability of GPi/SNr to inhibit the thalamus, and also promoting movement. The SNc in the substantia nigra dense part of the midbrain releases dopamine, which excites the direct pathway and inhibits the indirect pathway, respectively, making it easier for motor signals to pass through. The balancing of this complex circuit ensures the proper execution of the motion commands. In Parkinson's disease, the dopamine deficiency of SNc leads to the weakening of the direct pathway and the enhancement of the indirect pathway, and finally the GPi/SNr over-inhibits the thalamus, making it difficult to initiate movement. In addition, STN over-discharge due to the loss of sufficient inhibition of GPe also strengthened the inhibition of the thalamus by GPi/SNr. This sequence of alterations can be quantitatively described using circuit models and results in abnormal β oscillations in the basal ganglia circuit, corresponding to clinical motor stiffness and tremor.

To visually depict this circuit, we illustrate the flow of information from the basal ganglia-thalamic-cortical circuit in normal and pathological conditions in the figure above. The green arrows represent excitatory signals (glutamate pathway), the red arrows represent inhibitory signals (GABA pathway), and the blue arrows represent dopamine pathways (which have a dual effect on direct/indirect pathways). It can be seen that in Parkinson's disease (decreased dopamine), the inhibitory effect of GPi/SNr on the thalamus is too strongly activated by the indirect pathway (red), resulting in a decrease in the excitatory signal from the thalamus back to the cortex, and the movement cannot be initiated smoothly.

In the simulation platform, we will build a mathematical model of the above loops. Considering the real-time and controllable nature of the requirements, we can use neural mass models or mesoscopic network models to approximate the overall dynamics of individual nuclei. For example, each structure (striatum, STN, GPi, etc.) represents the evolution of its average discharge rate with a set of differential equations that can fit actual physiological data (e.g., patient DBS electrode records). In particular, we introduce dopamine concentration as a key parameter of the model to modulate the gain of direct and indirect pathways. This allows for a numerical description of "dopamine reduction", the core lesion of Parkinson's disease. Previous studies have used a similar method to simulate the dynamic differences of brain networks before and after L-Dopa administration, and successfully verified the effectiveness of the model by comparing the simulation results with the patients' EEG data. We will draw on these results to enable the model to produce both neural activity in a normal state and the characteristics of the pathological state by adjusting dopamine parameters. On this basis, the subconscious subsystem of the ACPU can continuously run the loop model to generate a real-time brain activity data stream of virtual patients.

Further, we plan to expand the scope of simulation and integrate other related networks of the brain to build a more complete "virtual brain". Movement disorders not only affect motor circuits, but also involve cognitive and emotional circuits. For example, patients with Parkinson's disease often have cognitive decline (involving the hippocampus, cortico-striatal circuits) and mood disturbances (involving the limbic system). As a result, our platform modularly incorporates cortex-striatum-hippocampal circuits (mimicking cognitive functions, such as the working memory circuit) and limbic circuits (such as the nucleus accumbens, dopamine reward pathways, which are involved in depression and motivational disorders). These modules are interconnected with motor loops to further enhance the physiological realism of virtual patients. At the same time, it also provides a means to study Parkinson's comorbidities such as Parkinson's dementia. This is where the architecture of the ACPU comes into play: because the virtual brain model contains many circuits and a large number of neurons, the subconscious subsystem (the brain-like computing part) can advance the numerical solutions of these differential equations in parallel and efficiently, or run event-based simulations of spiking neural networks. The consciousness subsystem can monitor these simulation outputs and combine them with the DIKWP semantic model to understand and intervene in the simulation results. For example, when a simulation exhibits an abnormal pattern (e.g., β oscillations are too strong, the simulator exhibits difficulty in locomotion), the consciousness subsystem elevates this pattern to information/knowledge semantics ("increased motor inhibition" is detected) and triggers an intervention at the target level (e.g., "reduce β oscillations to improve movement"). Next, it can intervene in the "virtual brain" in the simulation by calling the control interface in the platform. This is exactly what the name of the platform means by "regulation": not only to simulate disease circuits, but also to intervene in them and observe the results.

The platform will provide simulation interfaces for a variety of interventions: including electrical stimulation (e.g., DBS electrode stimulation of a nucleus), drug effects (altering virtual dopamine levels or adding pharmacological effects on specific receptors), genetic/molecular manipulation (mimicking the effects of gene knockout or protein aggregation on neurons), and sensory**/motor training**(Adapting sensory input given by the virtual environment or asking a simulated patient to perform a specific task to observe neuroplastic changes). For example, a DBS stimulation module can parametrically set the position of the stimulation electrode (STN or GPi, etc.), the stimulation waveform (frequency, pulse width, voltage), the stimulation strategy (constant stimulation or adaptive closed-loop stimulation), etc., and then act on the corresponding part of the simulation network to change its kinetic equation in real time (e.g. adding periodic forcing terms or changing thresholds to the stimulated nuclei). Another example is a drug module that can act on dopamine parameters or specific ion channel parameters to simulate the kinetics of the drug (e.g., levodopa works by gradually increasing dopamine levels, but fluctuating over a long period of time). Through these interfaces, we are able to try various treatment options in a virtual environment. The consciousness subsystem of the ACPU intelligently selects and combines these interventions according to the principles of the DIKWP Wisdom layer. For example, it can mimic a combination of drugs followed by electrical stimulation, compare the effects of different frequencies of DBS, or even try novel interventions (e.g., on-demand triggered adaptive stimulation, stimulation strategies that work in synergy with the patient's purpose).

An important goal of the platform design is to achieve closed-loop control and adaptive simulation. That is, the simulation not only passively runs the preset parameters, but can be adjusted in real time based on the results halfway through. For example, we can have the ACPU act as a "virtual doctor", constantly monitoring the key indicators of the virtual patient (such as motor function score, network synchronization level, etc.) during the simulation process, and making decisions to adjust the intervention parameters based on the DIKWP model, just as the clinical doctor adjusts the medication or adjusts the intensity of the stimulus according to the patient's response. Let's take a specific scenario: at the beginning of the simulation, the virtual patient is in the "off phase" (symptoms are severe, dopamine is low), and the ACPU chooses to give the drug first (simulating a dose of levodopa) according to the goal (the Purpose layer wants the symptoms to be halved). After a few minutes of simulation, a significant decrease in the "basal ganglia β oscillation" was detected, and the motor index improved, but at the same time there was an excessive γ oscillation (a side effect signal, which may correspond to dyskinesia); At this time, the ACPU consciousness subsystem judged the drug overdose according to the rules of the knowledge layer, and then adjusted the strategy: reduce the drug dose and turn on the DBS stimulation at the STN site to stabilize the network rhythm, and strive to maintain the improvement of movement while inhibiting abnormal oscillations. In this way, the platform completes an adaptive intervention simulation loop. Through a large number of such simulation experiments, we can find the optimal treatment plan, such as when to adjust the drug, when to add DBS, what mode to use for DBS, and so on. In fact, in reality, adaptive DBS (aDBS) has shown significant results over conventional constant stimulation: the use of implanted electrodes to monitor the patient's nerve signals in real time and adjust the stimulation as needed can reduce the duration of Parkinson's symptoms by up to 50%. Our simulation platform will provide a development and testing environment for these closed-loop therapies. For example, the effects of different sources of feedback signals (cortical EEG or STN local field potentials) and different control algorithms (threshold control or machine learning prediction) on the efficacy of aDBS can be tested on virtual patients to help design better aDBS strategies.

It is worth noting that the platform will not only serve the research objectives of this project, but will also become an open research tool in its own right. The architecture is scalable: researchers can add new models of brain regions as needed, substitute parameters to mimic other movement disorders (e.g., Huntington's disease, where striatal GABAergic neurons are lost and can be changed), or for other areas of brain science research. Through this engineering path, the platform output can continue to be iteratively upgraded. We will also establish an evaluation system for the platform, including simulation accuracy (how well the simulation results match real physiological data), real-time performance (simulation acceleration ratio), and performance evaluation indicators for intervention program optimization. For example, statistical methods were used to compare the improvement of virtual patients' symptom scores under different intervention regimens as a basis for evaluating the advantages and disadvantages of the regimens. Another example is to evaluate the credibility of the platform by verifying whether the platform can reproduce known experimental phenomena (such as the effect of radiofrequency burning surgery on tremor in the basal ganglia in the dopamine deficiency model) based on the literature data. These metrics will help us to continuously improve the model to make it more biologically realistic and predictive.

In summary, the ACPU-driven brain-like circuit regulation simulation platform proposed in this section is a key bridge from theory to application in this project. It "lands" the abstract DIKWP semantic model into a working virtual brain, allowing us to repeat experiments in the computer that are limited by ethics, cost or time in reality, and accelerate the discovery of the best intervention strategies. At the same time, in the development process of this platform, we will continue to combine the latest data from neuroscience (such as large-scale brain network connection maps, human brain stimulation response data, etc.) and the latest artificial intelligence technologies (such as more realistic neuron models and more efficient parallel computing chips) to ensure that the platform has sustainable evolution capabilities and will not stay in the same version. This is also in line with the requirements of the National Science and Technology Major Project for "platform construction": to establish an open and expandable platform to serve more research in the future.

Parkinson's Disease Mechanism Reconstruction and Drug Testing Environment Supported by Virtual Patient System

In the previous section, we described a brain-like circuit simulation platform that is equivalent to building a model of the "brain of disease". This section focuses on the construction of "virtual patient" systems, i.e., how simulations can be extended to the patient level for reconstructing disease mechanisms and conducting drug testing. The virtual patient should include not only the cerebral circuits, but also other physiological and pathological processes related to the disease, as well as external clinical manifestations and environmental interactions. In this way, we can fully reproduce the disease process of a person with Parkinson's disease in the digital space, from molecular, cellular to behavioral, for research and testing of new therapies.

There are several key elements to a virtual patient system:

1. Integration of multi-scale mechanisms: The pathological mechanisms of Parkinson's disease span multiple scales: at the molecular level, there are α-synuclein aggregation, mitochondrial dysfunction and neuroinflammation; At the cellular level, there is neuronal degeneration in specific brain regions; abnormal neural network dynamics (e.g., oscillation) at the loop level; At the systemic level, there are motor, cognitive, and emotional dysfunction. This virtual patient will integrate these levels of mechanistic models. For example, at the molecular-cellular layer, we can use a dynamic causal model based on literature to simulate the loss of dopamine neurons, including how age, genetics, and environmental factors lead to protein aggregation and neuroinflammation, which in turn triggers apoptosis or autophagy imbalance. This model can describe the proportion of neurons surviving over time and the effects of drugs on it (e.g., certain neuroprotective agents slowing apoptosis) using differential equations or state shift models. These micro-level states affect the parameters of the macro-loop model (e.g., dopamine levels), forming an upstream and downstream interface. Similarly, peripheral physiological models, such as musculoskeletal models, can be added to simulate motor execution and feedback, so that the virtual patient can "move" and their motor performance can be presented by simulating a simple anthropodynamic model (e.g., the handstand pendulum model for gait stability and the joint drive model for motion amplitude). Through multi-scale coupling, we reconstruct the chain of disease from pathogenic agents to dysfunction.

2. Individualized parameter setting: Each real patient has different symptoms and disease course, and the virtual patient system should also support parameterized individualization. We will design a series of parameters to represent the individual characteristics of the patient: such as the stage of the disease (early, intermediate, late), the type of predominant symptoms (predominantly tremor or stiffness), comorbidities (presence or absence of cognitive impairment, depression), age and gender, genetic background (type of mutation), etc. These parameters govern the settings inside the model. For example, the early PD corresponds to a low rate of neuronal loss and a certain level of residual dopamine. The tremor-dominant type can enhance the rhythm nodes associated with tremor occurrence in the loop model. In combination with depression, changes in the 5-HTergic system can be added to the model. By adjusting the parameters, we can "generate" different virtual patients. Ideally, we can customize the virtual patient based on the data of the real patient to achieve a "digital twin": the data of the individual's brain image, EEG, clinical scales and other data are mapped to the model parameters, so that the model can simulate the pathological characteristics of the patient. This will help personalize medicine in the long run.

3. Virtual Experimental Environment: One of the purposes of building a virtual patient system is to test drugs and therapies in a risk-free environment. Therefore, we need to design a virtual experimental process: various interventions can be applied to the virtual patient and the multi-dimensional output can be observed. Outputs include simulated clinical symptoms (e.g., motor performance scores inferred from motor loop activity, smoothness of virtual gait, virtual grip strength, etc.), as well as indicators of internal mechanisms (neuronal survival, protein aggregation levels, etc.). We will develop a visual interface to present these outputs and provide a statistical analysis module to compare different experimental conditions. Researchers were able to design experiments with virtual patients as they would in real experiments, such as testing hypotheses about "Can drugs that inhibit protein aggregation slow neuronal death and thus improve long-term motor function?" We can add the effect of the drug on the protein aggregation process to the model (set its effect on reducing the aggregation rate based on in vitro experimental data), and then let the virtual patients start to simulate the 5-year disease course from middle age, one group administers the drug and the other does not intervene, and observe whether the difference in neuronal survival and exercise capacity indicators between the two groups is significant after 5 years. If the virtual trial shows a positive effect, consider the feasibility of a real clinical trial. Similarly, we can test various drug combinations and intervention times at high throughput. Because there are no ethical and safety limits to virtual experiments, we can even try radical scenarios such as, "If I take a drug regularly for 10 years before the onset of symptoms, will I almost be able to avoid the onset of the disease?" "This kind of problem that cannot be immediately confirmed in reality can be predicted by the model.

4. Drug Screening and Repurposing: Drug screening can be significantly accelerated with a virtual patient system. We will integrate a drug database containing drugs known to target Parkinson's disease and related pathways and their mechanisms of action (receptor affinity, effects on pathway nodes, etc.). Combined with AI technology, we can have the system automatically search for candidates in the drug space that have an ameliorating effect on specific mechanism abnormalities. For example, if a model indicates that a patient's disease progression is primarily driven by an "inflammatory cascade," the system can screen the database for anti-inflammatory or immunomodulatory drugs that significantly reduce inflammatory mediator levels and neuronal death in the model, and recommend advancing the drug to animal trials or clinical trials. This computer-aided approach to drug repurposing is expected to find new uses from existing drugs. In fact, there have been attempts to use computational models to discover Parkinson's disease-related metabolites and potential drug targets. Our system will go one step further and provide a more intuitive basis for evaluating drug effects directly in full-course simulations. In addition, for new drug molecules, we can also interface with molecular dynamics and QSAR models to preliminarily evaluate their ability to cross the blood-brain barrier and target binding, so as to improve the screening efficiency.

5. Validation & Calibration: In order to guarantee the reliability of the virtual patient, it is necessary to perform sufficient verification calibration. Data sources include published clinical statistics, cohort study data, and data from model animal experiments that may be performed by our collaborating team. We will compare the model output with real-world data, e.g., whether the model predicts whether the average decrease in motor scores in untreated PD patients after 5 years is consistent with epidemiological observations; Whether the model simulation of "using a drug can extend the life of neurons by X%" is consistent with the results of animal experiments. If there is a deviation, we go back to the DIKWP semantic network to find where the mechanism may be missing or the parameters are wrong, and then fix the model. For example, if the model does not exhibit the motor fluctuations that are common in some patients, it may indicate that we have not included in the model the mechanisms of sensitivity changes (e.g., peak-trough effects) caused by prolonged use of levodopa, in which case a related mechanism module or adjustment parameter curve can be added. Through continuous iteration, we strive to make the model have the ability to predict quantitatively, that is, given an intervention plan, the model can approximate the trend of the outcome.

By constructing a virtual patient system, we open up a new dimension for the study of movement disorders such as Parkinson's disease: from qualitative description to quantitative reconstruction. Researchers are able to directly "see" the invisible parts of the disease process (such as the rate of cell death) in the virtual patient, and can arbitrarily manipulate various factors to test the hypothesis. This will greatly accelerate our understanding of the pathogenic mechanisms of disease. For example, in a virtual patient, it is possible to split the contributions of the various pathways: simply to see if the α-synuclein aggregation itself can cause disease, or does it have to superimpose aging/inflammatory factors? This helps to unravel the causal relationship between the entanglement of various factors. At the same time, for treatment development, virtual patients are low-cost test beds. While drug development has traditionally been a lengthy process of cell experiments, animal experiments, small-scale clinical trials, and scale-up clinical trials, virtual trials can simulate a multi-year process in a matter of hours, so as to eliminate apparently ineffective or serious side effects in advance, and focus valuable resources in the most promising directions. Of course, virtual trials cannot completely replace real clinical trials, but they can greatly improve the efficiency and success rate of R&D, and have proven to be valuable in oncology and other fields. Through the exploration of this project, it is expected to establish independent digital disease modeling and drug screening capabilities, and reduce the dependence on foreign experimental data and experience.

Overall, the Virtual Patient System pushes this project into the era of digital twins of disease mechanisms and intervention research: we can not only digitally reproduce diseases, but also digitally try treatments, which is no less important than inserting "second eyes" and "second hands" for medical research. This lays the foundation for true proactive medicine, because only by fully understanding the full picture of the disease and being able to predict the effect of interventions can we proactively and preemptively change the trajectory of the disease, rather than passively waiting for the disease to evolve and then catching up with the treatment.

Research on brain-computer interface and motor purpose remodeling mechanism

Brain-Computer Interface (BCI) technology is a rapidly developing hotspot in the field of neural engineering, which directly reads and decodes the neural activity of the brain to control external devices or computers, so as to achieve direct communication between the brain and the outside world. In the intervention of movement disorders, BCI has a unique application potential: on the one hand, it can read the patient's residual motor purpose , bypass the damaged neural pathway and directly drive the assistive device to help the patient complete the movement; On the other hand, feedback signals or stimuli can also be fed into the brain, combined with the patient's purpose, to enhance or correct their motor output. This project integrates brain-computer interface technology into the active medical intervention system, and explores the mechanism and method of motor purpose reconstruction, making it a bridge between patients and intelligent intervention platforms.

1. Signal acquisition and decoding of motor purpose: For patients with Parkinson's disease, due to the dysfunction of the basal ganglia circuit, when they try to exercise, the motor areas of the cerebral cortex still send motor command signals, but these signals are impaired during transmission (such as over-inhibited thalamic filtering) and fail to reach the spinal cord effectively. BCI captures these brain signals and provides us with a window into the patient's true purpose. Signals available include invasive cortical electrode array recordings (e.g., Utah microelectrodes, which can record single neuron discharges or local field potentials), deep brain electrodes (e.g., DBS electrodes implanted with STNs can also be used for recording), and non-invasive electroencephalography (EEG) or magnetoencephalography (MEG). Clinically, fully invasive options have been limited, but for patients with severe movement disorders, some attempts have emerged, such as cortical surface ECoG electrodes combined with BCI to control the exoskeleton. We will track signals that can be acquired in conjunction with the patient's therapy (e.g., the DBS device can be stimulated and recorded simultaneously after the upgrade to a bidirectional device), as well as a wearable, high-density EEG device to obtain motion-related signals.

Decoding the motor purpose requires translating these neural signals into meaningful instructions. Machine learning and deep learning techniques, such as training convolutional neural networks or time-frequency analysis algorithms, can be applied here to identify features of different motor imaginations or attempts (e.g., patients trying to raise their arms, legs, etc.) from cortical signal patterns. Previous studies have shown that AI models can distinguish brain MRI patterns between Parkinson's patients and healthy people, and even use EEG to predict exercise status. We collect a certain amount of data to train the decoder. For example, in a simulated or real-world setting, patients are asked to imagine/try a number of simple actions while recording the signal, and supervised learning is used to establish a mapping of the signal pattern to the purpose. Once trained, our system can monitor the patient's brain activity in real time and capture this purpose when they are trying to move but are blocked.

2. Refactoring and Execution of Motion Purpose: Obtaining Purpose is only the first step, and it is more important to convert Purpose into actual action output. This can be done through a variety of pathways: first, peripheral muscle stimulation or exoskeleton. For example, when BCI detects that the patient has the purpose of "lifting the right foot" but is unable to do it on its own, it can trigger an exoskeleton robot or functional electrical stimulation (FES) device attached to the leg to complete the foot lifting action on its behalf. Second, brain/spinal cord stimulation. If we can send the motor purpose back to the center so that other backup pathways in the brain or spinal cord can directly receive this instruction, it may also restore some motor function. For example, in some studies, scientists have realized the brain-computer-spinal cord interface, bypassing the injured segment in a person with spinal cord injury, and electrically stimulating the corresponding segment of the spinal cord after decoding the motor cortex signal through the computer, so that the paralyzed limb can move again. For Parkinson's patients, a similar idea can be envisaged: for example, the motor cortex signal is decoded and directly stimulated downstream of the patient's motor nerve pathway (e.g., stimulation of anterior horn neurons of the spinal cord) to achieve direct control bypassing the diseased basal ganglia. This, of course, requires addressing the issue of coordination and precise control. Our simulation platform can help test the feasibility and effectiveness of this type of motion purpose refactoring.

3. Convergence of BCI and ACPU/DIKWP system: In the platform architecture of this project, BCI is both an input device and an output device. On the one hand, BCI can be used as an input to inject the real Purpose data of the patient's brain into our DIKWP model, so that the Purpose layer and the Wisdom layer of the virtual patient can obtain a reference for the patient's subjective goals. For example, if a patient wants to walk but can't, this purpose can be captured by the system and considered in the Wisdom layer (e.g., the system will be more active in looking for interventions to improve walking). On the other hand, BCI is an output that can directly feed the system's decisions back to the patient's brain or body to achieve human-machine integration. For example, when the system decides to "increase the stride length of the left limb now", the patient's cerebral cortex can be guided by brain stimulation to produce corresponding motion images to enhance the patient's perception of the amplitude of movement, or the exoskeleton can be used to allow the patient to experience the movement of a larger stride length to train the brain to re-establish a normal motor rhythm. This form of integrating AI decision-making into the patient's neural circuit can be seen as a "hybrid closed loop": humans and machines form a closed-loop system, where humans provide will, and machines provide assisted decision-making and motivational execution, and the two influence each other. In order to achieve such a fusion, we will investigate the role of self-explanatory mechanisms in human-computer interaction. For example, the system senses that the patient is repeatedly trying an action through the BCI and explains to the patient on the interface (or by voice): "I detect that you want to stand up but your legs are weak, I will activate the lower limb exoskeleton to help you stand up" and execute it with the patient's consent or acquiescence. This interpretation and feedback can enhance the patient's trust and cooperation with BCI adjuncts and reduce false triggering.

4. Learning and adaptation for motor purpose reconstruction: Brain-computer interface systems often require continuous learning and adaptation, as the patient's brain signal signature may change as the disease progresses or is trained. In this project, we will introduce an online learning algorithm into the subliminal part of the ACPU to continuously optimize the decoder based on feedback. For example, when BCI controls the exoskeleton to make the patient walk, the system can obtain the quality index of gait from the sensor, combined with the patient's subjective feedback, and adjust the decoding parameters through reinforcement learning, so that the mapping from purpose to action is smoother. This is similar to having a virtual patient and a real patient form a co-evolutionary twin system: the virtual end adjusts to the real person, the real end recovers under virtual guidance, and the two progress together. The structure of the DIKWP model provides a clear hierarchy here: the data/information layer corresponds to specific signals and actions, the knowledge layer can record knowledge such as "the user's Purpose X corresponds to the error in decoding Y", the Wisdom layer formulates adjustment strategies (e.g., prompting the user to change the way of imagining slightly, or automatically adjusts parameters), and the Purpose layer is always aimed at the final rehabilitation goal. This structured adaptation ensures that the learning process is traceable, safe and controllable, and does not change itself unpurposefully (which is important in medical care, as algorithmic changes can be dangerous).

5. Clinical application prospects: The research on brain-computer interface and motor purpose reconstruction will eventually produce several prototypes of practical systems. For example, the BCI assisted walking system: the wearable EEG or ECoG collects signals, and AI decodes the walking purpose in real time to control the exoskeleton of the lower limbs, so as to achieve smooth stride when Parkinson's patients are freezing of gait. Purpose-driven smart mobility aid: Automatically assists and stabilizes balance when a patient is about to stand to prevent falls. Speech/Writing Purpose BCI: To help patients with severely limited movement but clear consciousness communicate with the outside world through brain-computer communication. With the miniaturization of hardware and the advancement of algorithms, these BCI devices are expected to become practical. In the longer term, it may also be combined with a deep brain stimulation device to form a "two-way brain-computer interface": it not only provides therapeutic stimulation, but also reads signals to assist the patient in movement or adjust stimulation parameters. It is foreseeable that in the next 5-10 years, the application of BCI in the field of movement disorders will continue to expand, and even become an important rehabilitation tool like pacemakers and prostheses. Through the forward-looking research of this project, we can enable China to seize certain technological opportunities in this field, especially the Explainable BCI combined with the DIKWP artificial consciousness model, which will be more competitive and safe.

All in all, the brain-computer interface and motor purpose reconstruction has injected the soul of "human-centered" into the active medical intervention system: it ensures that no matter how intelligent the AI is, it is centered on the subjective wishes and needs of the patient. This realizes a real closed loop of human-machine cognition, so that technology is no longer a cold external intervention, but an extension and supplement of patients' capabilities. In the management of diseases such as Parkinson's disease, such technologies have the potential to re-empower patients – even if their physical abilities are impaired, they can still actively participate in and control the recovery process through brain-computer interfaces, rather than being completely reactive in treatment. This shift is precisely the ideal of active medicine: Patient actively engaged in therapy with AI assistance, rather than passive recipient of care.

Domestic active medical intervention program, drug screening and intervention device design

The ultimate goal of this project is to transform the above theoretical models and technical achievements into practical active medical intervention programs and complete sets of devices, and provide domestic innovative solutions for movement disorder neurological diseases such as Parkinson's disease. This includes a complete process from prevention and diagnosis to treatment and rehabilitation, as well as the core hardware and software systems that support the process. In this section, we will synthesize the previous research, delineate the outlines of these schemes and devices, and emphasize their engineering realization paths and localization significance.

1. Comprehensive intervention program design of active medicine: Based on the DIKWP model and the concept of active medicine, we design the intervention program throughout the whole course of the disease, reflecting the idea of "early detection, early intervention, and full management". These include:

• Early Screening and Prevention: Using virtual patient models, we have identified some of the early digital biomarkers of Parkinson's disease (e.g., decreased sense of smell, abnormal sleep behavior, subtle bradykinesia, etc.) as well as potentially high-risk individuals (e.g., people with certain susceptibility genes or a history of pesticide exposure). The plan proposes to establish a community-level screening program, using simple and accessible domestic tools (olfactory test kits, smartphone apps for exercise tests, wearable devices to monitor sleep and gait, etc.), combined with AI evaluation models, to conduct risk assessment for middle-aged and elderly people. Once a high risk is identified, it will be included in the active intervention track immediately after being clinically confirmed. For example, lifestyle interventions (exercise therapy, Mediterranean diet, etc.) combined with neuroprotective drugs (eg, serotonin agonists, glucagon-like peptide-1 receptor agonists, etc.) are used in early-stage patients or high-risk patients. Here, the DIKWP Wisdom layer takes into account the adherence and cost-effectiveness of the intervention and tailors an individualized approach (e.g., an exercise program that does not interfere with daily routines for people who are still working).

• Proactive treatment after symptom onset: For patients who have developed motor symptoms, we are no longer limited to the traditional "step-by-step medication" and are introducing intelligent decision support. Specifically, the platform of this project is used to build a digital twin model for each patient, input their specific disease parameters, and the virtual patient system simulates the response to various drugs and dose combinations, so as to assist doctors in selecting the best regimen. For example, for patients with medium-term fluctuations in drug efficacy, the effects and side effects of adding a COMT inhibitor or an MAO-B inhibitor can be tested in the model, and the optimal combination can be recommended. This decision support will be provided to clinicians in the form of reports or interactive interfaces to help them make more informed adjustments to their treatment. For DBS surgery, we can also use the model to predict whether a patient is more suitable for STN or GPi stimulation (because the effect of the two targets is slightly different), as well as to predict the possible postoperative improvement for reference. Throughout the process, the systematic self-explanatory mechanism will clearly present the recommendation basis to the doctor to ensure that the clinical decision-making is transparent and credible.

• Closed-loop monitoring and dynamic intervention: We design a home remote monitoring and intervention system, in which patients wear domestic multi-modal sensing devices (such as wearable motion sensors, watches to detect tremor and motion, and smart cameras at home to analyze gait), and the system obtains data streams every day and generates DIKWP data-information reports through AI analysis, such as "today's tremor is 20% higher than yesterday, and the decrease in the number of turns at night may indicate poor sleep". This information is reported to the cloud platform, and the knowledge layer judges that there may be problems such as insufficient dosage or side effects, and the Wisdom layer formulates an adjustment plan, such as "It is recommended to increase the drug dose by half a tablet tonight, or turn on the sleep physiotherapy mode". The system can automatically notify patients or families and automatically perform some interventions when authorized (e.g., smart pill boxes automatically adjust the amount of medication released at night). This closed-loop system uses the domestic Internet of Things and 5G technology to realize remote program control: as the current domestic DBS already supports, the implanter can be programmed through the network. Doctors can also view patient status and intervene remotely through the platform. This allows patients to receive ongoing care outside of the hospital and no longer rely solely on regular follow-up appointments to adjust treatment.

• Rehabilitation and retraining: In addition to medication and surgery, the place of rehabilitation in active medicine is particularly emphasized in our program. Using the project's BCI and exoskeleton technology, we can provide personalized rehabilitation exercises: for example, with the help of Purpose Recognition, let the patient lead some training games in virtual reality to exercise balance and coordination; Or use robot-assisted walking while adjusting the pace to the patient's real-time purpose to reshape their neural pathways. Rehabilitation data are also fed into the DIKWP model to help refine the assessment of patient function and the selection of priorities for subsequent interventions (e.g., if a patient is found to have a slower recovery of fine motor hand, the Wisdom layer will adjust the protocol to enhance hand rehabilitation). The whole program is centered on the patient's purpose and quality of life, and is improved dynamically and cyclically, so as to truly achieve active, whole-process and individualized intervention.

2. Modernization platform for drug screening and R&D: The drug screening part of this project will not only stay at the model level, but also output some candidates and R&D clues with drug potential. We will connect the potentially effective drugs found in virtual screening with domestic pharmaceutical research teams to carry out in vitro and animal validation. This is expected to give rise to the secondary development of domestic independent innovative drugs or new uses of old drugs. For example, if the model suggests that a Chinese medicine compound that has been marketed in China has a potential effect on the alleviation of Parkinson's disease, we can quickly organize preclinical and clinical trials to obtain approval for the new indication in a short period of time. For another example, for the screened new target molecules, we can coordinate the domestic new drug R&D team to design the lead compound. The support of this program can cover early-stage R&D and increase the success rate. Within the 3-5 year period of reporting, we expect to screen at least a few drug candidates for further study and develop new patents for Parkinson's disease to protect or improve symptoms. Through such efforts, China is expected to gradually narrow the gap with the international advanced level in Parkinson's disease drugs, and realize the shift from mainly relying on imported drugs to drugs with independent intellectual property rights.

3. Localization design of intervention devices: In terms of hardware devices, the domestic equipment we plan to develop/improve includes:

• A new generation of deep brain stimulation (DBS) system: Based on the results of this project, a domestic brain pacemaker with adaptive stimulation function was designed. The specific features are: implanted electrodes to increase sensors, real-time recording of local EEG, and built-in AI algorithms (which can be based on the simplified ACPU principle) to adjust stimulation parameters in real time. The device realizes the main computing through domestic chips, and strives to reach the international level of similar products or even better. This project will provide the verification of algorithms and control logic, and the project implementation will be improved in cooperation with existing DBS vendors in China. At present, domestic manufacturers (such as Pinchi Medical) have launched perceptible, 3.0T NMR compatible, and remotely programmed advanced brain pacemakers, and we will further integrate closed-loop intelligence on this basis. After the launch of such a device, China will become a country that has mastered the core technology of adaptive DBS after the United States, and provide affordable high-end treatment for the majority of patients.

• Portable EEG/Magnetic Brain Monitoring and Stimulation All-in-One Machine: For early intervention and home rehabilitation, we design a non-invasive brain-computer interaction device. It consists of a multi-channel dry electrode EEG acquisition cap and a transcranial electromagnetic/magnetic stimulation module that can help patients with specific training or symptom control at home. For example, when a brain signal is detected that the patient has a motor block (frozen gait), the device automatically administers transcranial electrical stimulation in the motor cortex to try to break the abnormal brain rhythm and make the patient walk again; Or when the patient is depressed, a specific frequency of transcranial magnetic stimulation improves emotion. This device can be worn by the patient on the head, connected to the mobile phone APP through wireless, the doctor can set the program remotely, and the patient can use it according to the prompts. All components and software are domestically produced to ensure data security and supply chain security.

• Exoskeleton Rehabilitation Robot: As a key device at the output end of BCI, we will cooperate with domestic rehabilitation equipment manufacturers to transform the existing lower limb exoskeleton or training robot to make it compatible with brain-computer interface control. Its innovation lies in the addition of an AI control module, so that the robot can not only move according to a fixed program, but also respond to the patient's active purpose to adjust and cooperate. The structure adopts domestic high-performance lightweight materials and servo motors to ensure the comfort and safety of patients. The control algorithm is provided by us, especially the Wisdom layer concept of the DIKWP model into the control strategy, so that the robot can "know" the patient's state, such as actively reducing the assistance and prompting a rest when the patient is detected to be fatigue or resistance.

• Smart drug delivery devices: To enable more granular drug interventions, we are considering implantable or wearable drug delivery devices. Examples include implantable drug pumps, which are used to continuously and slowly release dopamine analogues or other drugs to specific sites (similar to current duodenal gel pumps, but we explore the possibility of intraventricular or targeted drug delivery). and a subcutaneous smart pill box, which can be used to inject drugs at regular daily intervals according to the program, or to increase the dose in time when the system detects a worsening of symptoms. Through these devices, drug therapy can also achieve a certain degree of closed loop. All devices will be designed in accordance with national medical device specifications and will be ready for registration.

4. Evaluation system and standard construction: In parallel with the rollout of these programs and devices, we will also develop an evaluation system to quantify their effectiveness and safety. This includes: efficacy indicators (e.g., percentage improvement in UPDRS score, change in quality of life questionnaire score), initiative indicators (patient engagement, improvement in self-management ability), safety indicators (incidence of adverse events, device reliability), etc. In addition, we will engage with regulatory and standards bodies to incorporate some of the practices of proactive medical intervention into industry standards. For example, guidelines for the integration of digital diagnosis and treatment of Parkinson's disease have been formulated, including digital biomarker screening and virtual model-assisted decision-making. Formulate safety certification standards for brain-computer interface products, etc. These efforts will help standardize and disseminate our results, paving the way for subsequent industrialization.

To sum up, this section depicts a blueprint for collaborative innovation of industry, academia, research and medicine in China: through the special support of the state, we have not only made theoretical and model breakthroughs, but also made the output really "visible and tangible".of programs and equipment to bring good news to patients. In particular, the emphasis on the word "domestic" means that the core technology and products are independent and controllable, and are not controlled by others. This is critical to ensure large-scale clinical applications and to cope with international competition. At the same time, we also expect these results to have the attribute of sustainable evolution: with more data accumulation and technological progress, the solution can be updated and iterative, and the device can also be upgraded (such as using more advanced ACPU chips in the future to improve processing speed and algorithm performance). All this will establish China's leading position in the field of active intervention for movement disorders and provide an example for active medical exploration of other chronic diseases.

Three-to-five-year project phased tasks and key outcome nodes

The project is planned to be implemented in phases over a period of 3 to 5 years, and gradually achieve the goal of moving from basic theoretical research to system integration to demonstration application. The main tasks and milestones for each phase are planned in chronological order below:

Phase 1 (the initial year of the project, about the first year): theoretical model and key technical research

• Task 1.1: Improve the DIKWP semantic model of dyskinesia Based on the literature and expert knowledge, the first version of the DIKWP semantic network for movement disorders such as Parkinson's disease was constructed, including the main concepts, hierarchical relationships and the definitions of 25 transformation modules. Key nodes: complete the design of the semantic network ontology and publish an internal technical report; Preliminary validation of the model can represent typical pathological processes and gain expert acceptance.

• Task 1.2: Prototype the ACPU architecture. Design the framework of the Artificial Consciousness Processing Unit (ACPU), including the interface definition of the subliminal LLM module and the conscious DIKWP module, and select the hardware implementation route (FPGA prototype or GPU simulation implementation). Key nodes: Complete the ACPU architecture design document, build a simulation environment, and run a dual-cycle awareness process demonstration in a simple scenario.

• Task 1.3: Modeling the dynamics of the basal ganglia loop. Establish a mathematical model (neuromass or large-scale neural network model) of Parkinson's disease-related basal ganglia-thalamic-cortical circuits to realize the mechanism by which dopamine levels regulate the activity of the pathway. Key node: Publish/submit an academic paper reporting the structure and parameters of the model, and verify that the model can reproduce the typical neural signal characteristics of Parkinson's disease (e.g., β oscillation) [36†L153-L160}.

• Task 1.4: Virtual Patient System Framework Development. Develop a software framework for the virtual patient system, integrate the interface of multi-scale sub-models, and build a basic data input and output pipeline. Key node: Complete the virtual patient v1.0 framework, so that it can load and run the basal ganglia model, and visually output motor symptom indicators.

• Task 1.5: Preliminary experiments with brain-computer interfaces. A number of volunteers or healthy subjects with Parkinson's disease were selected to carry out preliminary EEG/EMG acquisition experiments to lay the foundation for BCI decoding. Key node: Collect brain-computer training datasets, and initially train a simple model to recognize 1-2 kinds of sports purposes (with an accuracy rate of more than 70%).

Phase 1 Summary: This year we will establish the core theoretical and technical foundation of the project, including the DIKWP model, simulation model and prototype system, and provide tools for subsequent work. It is expected that several preliminary results will be achieved at the end of the phase, such as model papers, prototype demonstrations, and verification of the feasibility of the project roadmap.

Phase 2 (mid-project, approx. 2-3 years): System integration and functional improvement

• Task 2.1: Integrated development of active medicine platform. Gradually integrate the DIKWP model, virtual patient, ACPU, and BCI modules into a unified platform architecture to connect data flow and control flow. Focus on the development of semantic interpretation interface and decision control module. Key node: complete the alpha version of the "Active Intervention Medicine Platform", which can simulate a virtual patient in a laboratory environment and conduct closed-loop intervention trials (such as drug delivery simulation + stimulation simulation + BCI feedback).

• Task 2.2: Iteratively optimize the virtual patient model. According to the preliminary model of the first stage, the model parameters were optimized and the mechanism was supplemented. Introduce more physiological data to validate the model, such as using public datasets to calibrate the accuracy of the model in predicting symptoms. Key node: The model version has been upgraded to 2.0, which can accurately simulate the symptom curves of PD patients at different stages, compare the model prediction with clinical data, and submit the relevant results to high-level journals.

• Task 2.3: Virtual screening and candidate identification of new drugs. Large-scale drug/target screening calculations using well-established models. Work with the pharmacology team to analyze the screening results and select the 3-5 most promising drug candidates or combinations. Key nodes: Form a detailed virtual screening report, list the candidate interventions and mechanism of action predictions, and give follow-up experimental verification plans for at least 2 of them.

• Task 2.4: BCI Prototyping System Development. Development of BCI prototypes for motion purpose reconstruction, including portable capture devices and decoding software. In collaboration with the Department of Rehabilitation, the algorithm was tested on several patients and adjusted. Key node: The BCI system can decode the patient's simple movement purpose in real time (with a delay of <300ms) and drive the motion of the computer cursor or simple mechanical device, which has preliminarily proved its effectiveness.

• Task 2.5: Joint research of domestic devices. Carried out joint research and development with industrial partners such as brain pacemaker upgrade and exoskeleton control. The adaptive algorithm of intelligent DBS was verified by animal experiments (such as Parkinson's model rats), and the exoskeleton BCI was combined with healthy human experiments. Key nodes: Preliminary experimental data were obtained, such as the traditional mode of adaptive DBS in Parkinson's model animals to prolong the exercise time by X%, reduce the symptom score by Y%; The BCI exoskeleton improved gait uniformity by Z% in a patient with mild symptoms.

Phase 2 Summary: This phase brings together the various modules of the project into a more functional system and continuously optimizes it. The important results we expect to achieve during this period include: the completion of the prototype of the active medicine platform; The performance of the Parkinson's disease digital twin has reached usable levels; Discover a number of new cues for intervention; Basic verification of key technologies such as BCI and intelligent stimulation. Possible landmark results include: system integration demonstration (to demonstrate the platform's real-time simulation and intervention capabilities to experts and authorities), several authorized invention patents (such as adaptive DBS algorithm, BCI decoding method, etc.), and a series of papers published in international conferences or journals to establish the influence of the project in the academic community.

Phase 3 (late stage of the project, about 4-5 years): application verification and demonstration and promotion

• Task 3.1: Clinical trials and effectiveness evaluation. A Parkinson's disease center in a tertiary hospital was selected to conduct a small-scale clinical pilot. The content includes: using the platform to assist the actual patient's treatment decision-making (the doctor adjusts the plan according to the model recommendation) and observing the difference in efficacy compared with traditional empirical treatment; trial of the effect of BCI assistive devices in rehabilitation training; Remote monitoring and parameter tuning for patients receiving intelligent DBS. Key node: Complete the data collection of at least 30 patients in the controlled trial, and the results show that the patients who applied the project system have statistically significant improvements in symptom improvement, complication control and other indicators, and write clinical research papers or reports.

• Task 3.2: Industrialization Preparation and Regulatory Approval. Organize technical documents together with partner companies and prepare registration materials for related products. For example, the intelligent DBS system was submitted to the State Food and Drug Administration for review, and the BCI rehabilitation equipment applied for Class II medical device registration. Key nodes: Obtain the registration approval of at least 1 product or enter the green channel pilot application; The guidelines for active diagnosis and treatment of Parkinson's disease have been recognized by industry associations and have been published and applied within a certain range.

• Task 3.3: Training Outreach and Feedback Improvement. Compile user manuals and training materials, and train doctors and rehabilitation therapists in the use of the system; Select a number of hospitals to promote and trial the cloud-based decision support system. Collect user feedback to continuously improve system ease of use and functionality. Key nodes: 2-3 national training sessions will be held, and more than 50 professionals will master the use of the system; The system has been upgraded to the final version 3.0, with a friendly interface and stable operation, and has been tested to meet the requirements of medical information security.

• Task 3.4: Acceptance of project results and prospect planning. Comprehensively evaluate the completion of the project against the project indicators, and organize technical data and data archiving. Prepare for a larger-scale application promotion plan and possible commercialization paths, such as the establishment of a school-enterprise joint company to continue to advance. Key node: through the national project acceptance, the acceptance experts recognized the innovation and application value of the project; Formulate a plan for the next stage of development, and clarify how to promote this achievement nationally and internationally.

Phase 3 Summary: The final stage will really push our research into clinical practice and verify whether the "paper talk" translates into patient benefits. If it goes well, we will see real patients get better outcomes and improved quality of life because of our system. The hallmark of success at this stage lies not only in the data and papers, but also in clinical recognition and policy support: ideally, the system is endorsed and promoted by a team of leading Parkinson's experts in China, and health authorities see the promise of a proactive medicine model and provide policy support (e.g. inclusion in chronic disease management programs). We hope that by the end of the project, China has established a complete closed-loop management system for active intervention in Parkinson's disease in a demonstration center, and has the idea and foundation to promote this model to more diseases (Alzheimer's disease, motor neuron disease, etc.).

Milestones at a glance:

• Milestone 1 (about 0.5 years) :D IKWP cognitive model of movement disorders was preliminarily completed, the basal ganglia loop simulation model was built, and the internal workshop evaluation was passed.

• Milestone 2 (around the end of the first year): ACPU prototype and virtual patient v1.0 released, demonstrating the simple active intervention process, 2 patents submitted, and several papers published.

• Milestone 3 (approx. end of year 2): The integration test of the active medicine platform was successful, and the closed-loop simulation intervention was effectively operational; The performance indicators of the Parkinson's disease digital twin model meet the predetermined requirements; Candidate list generation.

• Milestone 4 (around the end of the 3rd year): BCI Purpose reconstitution system successfully validated (healthy subject experiments), intelligent DBS algorithm successfully validated in animals; The core system enters the clinical trial preparation stage.

• Milestone 5 (approx. 4-5 years): Clinical pilot data demonstrating significant improvement in efficacy; At least one smart device is approved for use; Project techniques and specifications began to be promoted in the industry.

Through the above-mentioned phased promotion and achievement output, this project will complete the whole process from theoretical innovation to clinical preliminary verification within 3-5 years, and effectively build an active medical intervention system for movement disorders integrating the DIKWP model. This not only achieves the goals required by the guidelines, but also lays a solid foundation for future development.

epilogue

The project of "Construction of Active Medical Research and Intervention System for Movement Disorder Neurological Diseases Integrating DIKWP Model" is based on solving the major public health problem of Parkinson's disease and other movement disorders, and has carried out systematic design and planning from basic theory, technology research and development to clinical application. We make full use of Professor Yucong Duan's DIKWP artificial consciousness model and its derivative theories, deeply integrate it with neuroscience, and innovatively propose a semantic hierarchical disease model and an artificial consciousness-driven simulation control platform. By introducing the ACPU architecture, semantic elastic network, and self-explanatory mechanism, our system has the ability to explain and self-regulate that traditional AI does not have. Focusing on this core, we have built modules such as virtual patients, brain-computer interfaces, and closed-loop intervention devices, forming an overall solution for active medicine.

The implementation of this project will fill the gaps of the existing technology in the following aspects: firstly, realize the digital reconstruction of the whole chain of Parkinson's disease and other diseases from data to purpose, which will help solve the mystery of the complex mechanism of movement disorders; Secondly, the international advanced adaptive closed-loop intervention technology, such as intelligent DBS and BCI-assisted exercise, has been developed, which significantly improves the treatment effect and individualization. Thirdly, a new paradigm of model-driven drug screening was established to accelerate the discovery of new therapies. Finally, an active health management system for the grassroots and families has been created to realize long-term companionship medical care for patients. All these innovations are expected to raise China's scientific and technological level in the field of movement disorder prevention and treatment to a new height.

Of course, the success of the project depends on the integration of multiple disciplines and the collaboration of multiple parties. In the course of the project, we will strengthen the collaborative research of experts in the fields of neurology, artificial intelligence, computer, biomedical engineering, etc., and closely cooperate with industry, academia, research and medicine. At the same time, we keep abreast of the latest international developments to ensure that our technology is always up-to-date and optimal, rather than working behind closed doors. The introduction of an evaluation system and a sustainable evolution mechanism ensures the reliability and vitality of the project results – our system will continue to improve itself with data and experience, becoming more intelligent and accurate. More importantly, the compound talent team cultivated and trained by this project will become an important force in the field of active medicine and artificial intelligence + medical care in China, providing human support for continuous technological iteration.

Looking to the future, if the system developed in this project is widely used, patients with Parkinson's disease will no longer rely on passive medication or waiting for surgery, but can receive individualized and comprehensive intervention at the early stage of the disease, and enjoy the protection and guidance of intelligent systems throughout the disease process. The quality of life of patients will be greatly improved, and the burden on families and society will be reduced. This reflects the purpose of science and technology to be people-oriented and serve people's livelihood. In addition, the methodology of this project can be extended to other chronic neurological diseases, creating a new era of disease digitalization and intervention initiative.

To sum up, this project will promote a comprehensive understanding of the mechanism of movement disorders in a scientific sense, create an independent intelligent medical platform in China in a technical sense, and bring new prevention and treatment methods in a clinical sense. Through 3-5 years of hard work, we are confident that we will deliver a high-quality answer sheet: to achieve the unity of theoretical innovation, technological breakthrough and clinical application, to achieve the goals guided by the guidelines, and to make significant contributions to the "research on the pathogenesis and intervention treatment of Parkinson's disease and other movement disorder neurological diseases" in China. At the same time, this project will also set a demonstration for the deep integration of artificial intelligence and medicine, and help China seize the commanding heights of future medical reform. We look forward to turning this forward-looking and socially valuable blueprint into reality with the support of major national science and technology projects, and bringing benefits to the majority of patients."