Background and significance

Spinal cord injury (SCI) and other central nervous system damage often lead to severe paralysis, and there are more than 1 million SCI patients in China and about 66,000 new cases every year. Patients with paralysis lose their motor function for a long time, which not only has a very low quality of life, but also brings a heavy burden to their families and society. How to effectively restore the motor function of paralyzed patients is a major challenge and urgent need for medical and engineering technology.

Status of traditional technologies: In recent years, breakthroughs in brain-computer interface (BCI) and neurostimulation have brought new hope to patients with paralysis. For example, patients with partial complete paralysis regain the ability to stand and walk by activating residual spinal nerve circuits through epidural electrical stimulation (EES). However, these recoveries mostly rely on pre-programmed stimulus sequences and require wearable sensors to detect residual movements to trigger, making the control not entirely "natural". Recent research has shown that the "brain-spine interface" that connects the purpose of brain movement directly to spinal cord stimulation allows patients to control lower limb movements in a more natural and active way. For example, a study implanted a 64-electrode array in the cerebral cortex of patients to collect motor purpose signals, and modulated the 16-electrode stimulation of the lumbosacral spinal cord in real time, establishing a digital bridge between the brain and the spinal cord, and achieving a breakthrough in the autonomic regulation of walking in paralyzed patients. These results demonstrate that closed-loop control of central signals and electrical stimulation is essential for the restoration of complex motor functions.

**Problems and bottlenecks: Although the above-mentioned brain-computer interface and spinal cord stimulation technologies are exciting, there are still many shortcomings: (1) Lack of high-level cognitive involvement: The existing system mainly focuses on signal acquisition and stimulation output, and lacks in-depth modeling of the formation of patients' motor purpose and its interaction with sensory feedback. After paralysis, due to the lack of actual motor feedback, the brain's ability to produce and regulate motor purpose will gradually weaken. The study found that the degree of desynchronization of the rhythmic rhythms (μ and β waves) of the sensorimotor cortex was significantly reduced when patients with chronic paralysis attempted to move the paralyzed limb, but there was no significant difference with those in able-bodied patients when purely imagining movement. This suggests that the disruption of the actual motor-sensory feedback loop leads to plasticity changes and dysfunction of the brain's motor purpose pathways, which need to be reshaped by specific training. Existing techniques rarely take into account the assessment and reconstruction of subjective motor awareness and purpose generation in patients with paralysis. (**2) Limited degree of closed-loop intelligence: Many systems use fixed stimulation parameters, which is difficult to adapt to changes in the patient's state in time, and is prone to neural adaptation and stimulation inertia. The lack of comprehensive perception and intelligent decision-making mechanism for multi-source information such as environmental changes, patient fatigue, and muscle response leads to limited control accuracy and collaborative efficiency. (3) **Insufficient central-peripheral coordination: It is often difficult for single spinal cord stimulation or muscle/nerve stimulation to take into account the reconstruction of central pathways and peripheral muscle activation. A "dual electrical stimulation" strategy has been proposed in the literature, which combines epidural spinal cord stimulation with muscle electrical stimulation to simulate the central feedforward signal and peripheral feedback signal, respectively, so as to more effectively reconstruct the spinal cord sensorimotor circuit. Animal experiments have shown that dual stimulation can simultaneously promote the structural and functional reconstruction of spinal cord pathways, activate genes related to pro-axon regeneration, and improve spinal cord neuronal excitability at 10–20 Hz. These suggests that central and peripheral synergistic stimulation is expected to be a new way to treat paralysis, but the research on such multiparametric synergistic control is still in its infancy. (**4) Multi-channel precise control and feedback: Restoring complex motor functions requires spatiotemporal and precise coordination stimulation of multiple joints and muscle groups, which involves electrode control of a large number of channels and processing feedback of massive data. At present, clinical spinal cord stimulators usually have only about 16 channels, and most of them lack the function of obtaining and real-time adjustment of sensory feedback signals, which is difficult to meet the needs of fine collaborative control.

New Theoretical and Technological Ideas: In view of the above challenges, it is urgent to introduce new cognitive theoretical frameworks and artificial intelligence technologies to guide the design and control of spinal cord-peripheral nerve interfaces, so as to realize the leap from "signal-driven" to "intelligent adaptive". The "Data-Information-Knowledge-Wisdom-Purpose" mesh cognitive model proposed by Professor Duan Yucong provides a unique foundation for this. The DIKWP model adds a fifth layer of purpose to the traditional "Data-Information-Knowledge-Wisdom" architecture, and replaces the one-way hierarchical structure with a nonlinear network structure. The semantic elements of each layer can be fed back and updated in both directions, so as to form a multi-level adaptive cognitive network. This new cognitive system is an academic milestone that can provide an interpretable and controllable semantic-level decision-making mechanism for complex AI systems. Crucially, the DIKWP model ensures that the AI's actions always point to the intended goal by embedding "Purpose" into the cognitive process, so that every decision step has clear human-understandable semantics. In the research of Artificial Consciousness (AC), Professor Duan's team further proposed the "DIKWP×DIKWP" architecture, which introduces the dual-cycle structure of basic cognitive processes and metacognitive cycles to achieve self-monitoring, self-reflection and self-regulation, which is regarded as an important path towards autonomous consciousness AI. The "dual circulation" artificial consciousness architecture gives the AI system a preliminary self-awareness and purpose-driven adjustment ability.

The above theoretical progress provides a new perspective for us to re-examine the reconstruction of motor function in paralyzed patients: we can regard the patient-environment-stimulus system as a whole containing data, information, knowledge, wisdom, and purpose, and restore motor control through artificial consciousness-driven dual closed-loop regulation. On the one hand, the DIKWP model is used to unify the representation and interactive modeling of five types of information flows, namely motor purpose, sensory feedback, nerve stimulation, muscle response, and environmental perception, which is expected to reveal the mechanism of central-peripheral information imbalance in the state of paralysis. On the other hand, the artificial consciousness system is introduced into rehabilitation training, and through virtual purpose feedback and adaptive stimulation, patients can re-"perceive" their own motor purpose to be recognized and realized, and rebuild the connection between the brain and the body.

Policy and application needs: This research direction is highly in line with the national science and technology strategy and clinical needs. "Brain science and brain-like research" has been listed as a key frontier area in the 14th Five-Year Plan, and brain-computer interface and neuromodulation technology are the focus of research, aiming to achieve functional reconstruction of major diseases such as paralysis。 For example, national and local action plans have been introduced to promote the completion of clinical trials of a number of implantable/non-implantable brain-computer interface products in the next few years to restore some of the language and motor functions of patients with aphasia and paralysis. Therefore, this project is carried out under the framework of the national guideline "Research on Spinal Cord-Peripheral Nerve Interface Technology", which not only responds to the direction of cutting-edge scientific and technological research, but also has great clinical translation value. By integrating the original DIKWP × DIKWP cognitive model and artificial consciousness theory, we will explore a new paradigm of intelligent rehabilitation, which is expected to break through the bottleneck of traditional technology and bring leapfrog functional recovery methods for paralyzed patients. This will not only significantly reduce the burden on patients' families and society, but also provide core technical support for China to seize the international commanding heights in the emerging industries of intelligent rehabilitation equipment and brain-computer interface.

In summary, this project has important research significance: academically, it will create a central-peripheral integrated cognitive interaction model to enrich new theories in the intersection of artificial intelligence and neuroscience; Technically, a multi-parametric and adaptive neuromodulation system should be developed to improve the rehabilitation effect of paralysis. In terms of application, it accelerates the implementation of the intelligent rehabilitation platform to benefit the majority of patients, in line with the strategic goal of "Healthy China 2030" and the task requirements of the national key research and development plan.

Research objectives and overall technical roadmap

The core idea of this project is to construct a closed-loop control system with a high degree of integration between the central and peripheral areas under the guidance of Professor Duan Yucong's original DIKWP×DIKWP network cognitive model and artificial consciousness (AC) theory, so as to realize the reconstruction and training of motor function in paralyzed patients. The specific research objectives are as follows:

1. Constructing a central-peripheral integrated DIKWP ×DIKWP cognitive interaction model: A network DIKWP cognitive interaction modeling system was established to integrate and represent five types of dynamic information flows**: motor purpose generation, sensory/perceptual feedback, neurostimulation signals, muscle group response, and environmental perception**——The interactive closed-loop mechanism. The model quantitatively describes the information transmission and feedback relationship between the central nervous system, peripheral nerves and the external environment, reveals the cognitive mechanism of motor control circuit dysregulation in the state of paralysis, and provides a semantic guidance framework for subsequent control strategies.

2. Analysis and reconstruction of the mechanism of motor purpose disorder based on artificial consciousness theory: Using the "dual circulation" structure in the artificial consciousness model, the metacognitive level is introduced to analyze the mechanism of motor purpose generation disorder and sensory feedback loss in paralyzed patients. To develop an artificial consciousness interaction training system for paralyzed patients, through virtual scenarios, biofeedback and other means, to simulate the closed loop of motor purpose-action-feedback, and reconstruct the generation and correction pathway of motor purpose in the patient's brain. The system is designed to train and repair the patient's brain motor purpose generation ability, enhance subjective initiative, and assist in reshaping the brain's control mapping of the body.

3. Precise spatiotemporal electrical stimulation control strategy for DIKWP hierarchical mapping: In the synergistic effect strategy of spinal epidural electrical stimulation and peripheral nerve/muscle electrical stimulation, the functional mapping method of each layer of the DIKWP model is introduced. Aiming at the functional goals pointed to by the motor Purpose layer, combined with the abstraction of the motor control mode by the knowledge and Wisdom layer, a multi-parameter spatiotemporal precision electrical stimulation scheme was formulated. These include: selecting the optimal stimulation site (posterior root nerve segment or peripheral nerve of the spinal cord), optimizing the timing and frequency of stimulation (according to the dynamics of the sensorimotor pathway, studies have shown that 10–20 Hz is an effective frequency band for rebuilding neural pathways), real-time adjustment of stimulation amplitude and pulse width, and coordination of the timing of central stimulation and muscle stimulation. A multi-parameter adaptive control algorithm was developed to enable the stimulation output to be adjusted in real time at the data/information layer according to the sensory feedback (such as electromyography, joint angle, force sensing, etc.) and the evaluation of the patient's state by the artificial consciousness system, so as to realize the refinement of different action purposes, and the spatial synchronization and time synchronization control of electrical stimulation and muscle contraction.

4. Prototype development of cognitive perceptual feedback multi-channel stimulation system: Design and development of a prototype of a high-density spinal-peripheral nerve stimulation system with more than 64 channels. The system will integrate cognitive perception feedback capabilities, that is, in addition to outputting multi-channel electrical stimulation, it will also synchronously collect multimodal information (such as electromyography signals, pressure sensing, accelerometers, etc.) from the patient's muscles, joints, and environment to form a closed-loop control. In terms of hardware, a programmable multi-channel stimulator has been developed, which supports the independent driving of more than 16 channels of epidural electrodes and several channels of peripheral nerve**/muscle electrodes, with millisecond-level timing accuracy and wireless communication functions. In terms of software, an embedded real-time control program was developed to allocate the stimulation parameters of each channel according to the strategy given by the DIKWP model. The prototype system will validate the ability to control at least the three major joints of the lower limb** (e.g., hip, knee, ankle) and their associated muscle groups. Through bench experiments, in vitro tissue or animal experiments, the muscle contraction effect, joint movement trajectory accuracy and safety of the system induced by different stimulation modes were evaluated, so as to provide a basis for subsequent human experiments.

5. Integrated intelligent rehabilitation training and evaluation platform: Build an intelligent rehabilitation platform that integrates the above achievements. The platform includes: spinal/peripheral multi-channel electrical stimulation subsystem, DIKWP structured function evaluation module, artificial consciousness human-computer interaction interface, and data management unit. Through the platform, personalized motor function recovery training for paralyzed patients is realized: the artificial consciousness interactive interface is used to present virtual training tasks to guide patients to produce motor purpose; The DIKWP evaluation module quantitatively evaluates the patient's current data, information, knowledge, Wisdom, and Purpose status (e.g., the data layer evaluates the EMG/EEG signals, the information layer evaluates the basic motor ability indicators, the knowledge layer evaluates the mastery of combined movements, the Wisdom layer evaluates the adaptability to environmental changes, and the Purpose layer evaluates the patient's active participation and subjective initiative). Based on the evaluation results, the parameters and training scheme of the electrical stimulation subsystem were adaptively adjusted. The platform will provide a doctor/therapist interface for setting training goals, monitoring the training process, and automatically recording and analyzing the patient's recovery progress. Finally, the platform is deployed in the clinical environment to implement personalized and intelligent motor function reconstruction and training services for paralyzed patients.

Objectives 1 and 2 provide the theoretical basis and software architecture of the project, objectives 3 and 4 focus on specific technology implementation and hardware development, and goal 5 focuses on system integration and application verification. In terms of the overall technical route layout, this project adopts the route of "theoretical model guidance, key technology research, prototype system development, clinical verification and application". Firstly, the DIKWP × DIKWP model were used throughout the whole process, and the motor purpose reconstruction and electrical stimulation strategies were studied under this framework. Then, the core technologies such as multi-parameter stimulation and feedback control were overcome, and software and hardware prototypes were developed. Finally, it is integrated into a complete platform that can be tested in clinical trials. The technical route is shown in the figure:

Fig.1 Conceptual diagram of the intelligent control system of the brain-spinal interface: a central-peripheral dual closed-loop control framework fusing the DIKWP ×DIKWP cognitive model. Purpose signals from the cerebral cortex are transmitted to spinal cord stimulation through the central circuit, and peripheral sensory feedback is transmitted back through environmental and somatic sensors. The artificial consciousness module (metacognitive ring) monitors and adjusts the entire process to achieve the voluntary recovery of motor function.

Figure 1 conceptually illustrates the intelligent control system to be constructed in this project: the patient's brain movement purpose is obtained by the implanted/non-implanted signal acquisition device, the purpose information is decoded by the wearable processing unit, and the DIKWP cognitive model is input for multi-level semantic understanding; Combined with the artificial consciousness module for self-monitoring, the processing unit generates dynamic adjustment instructions for the spinal cord/peripheral nerve stimulation program, and acts on the spinal cord and muscles through a multi-channel stimulation device to induce corresponding limb movements; At the same time, the muscle response and external environment information collected by various sensors enter the information/knowledge layer of the cognitive model through the feedback channel, which supports the system's evaluation of the current action effect and the optimization of the next decision. Metacognitive loops ensure that the system is adaptive, self-adjusting, e.g., autonomously changing the training rhythm and stimulation pattern when a patient's fatigue or ambiguity is detected. This "double closed-loop" structure that combines a central closed loop (brain-spinal cord/muscle) and a self-regulating closed loop corresponds to the dual-loop concept of the DIKWP × DIKWP model, which is expected to significantly improve the naturalness, effectiveness and safety of motor control in paralyzed patients.

Key technology research content and sub-module design

Focusing on the above overall goals, we subdivide the tasks into five modules, corresponding to five key technical directions. The research content and scientific and technical problems to be solved in each module are as follows:

(1) DIKWP× DIKWP cognitive interaction modeling module

Main contents: To establish a central-peripheral integrated DIKWP network cognitive model to simulate the interaction of different elements in the motor control system of paralyzed patients. This module first defines the connotation of five core elements in this research scenario: data D (such as the original signals of each sensor, including EEG/EMG, electrical stimulation potential, environmental signals, etc.), information I (useful information extracted from the data, such as exercise purpose signal characteristics, muscle contraction intensity, joint angle changes, etc.), and knowledge K(knowledge of human movement patterns, physiological laws, such as coordination patterns of walking, muscle recruitment sequences, balance maintenance strategies, etc.), Wisdom W (higher-level decision-making and strategies, such as the ability to adjust gait according to environmental requirements, adjust protocols based on training feedback), Purpose P(the patient's subjective expectation of a target action or function). Next, we will construct formal representation and association rules for these five elements, such as using directed graphs or tensor networks to represent dependencies and feedback paths between features. Based on the DIKWP brain region mapping theory proposed by Professor Duan Yucong, the above elements were mapped to the neuroanatomical structure and functional circuit of the brain-spinal cord-muscular system to form a model corresponding to cognition and physiology.

Innovation: Most of the traditional neural control models are one-way hierarchical reflection loops, and this module introduces the concept of DIKWP mesh bidirectional feedback, which regards motion control as a multi-level and multi-directional information flow network. For example, we will explore the reaction of the Purpose layer to the knowledge layer (the shaping of movement patterns by the patient's goals) and the modulation of the Purpose layer by sensory feedback (how successive failures/success experiences affect patient willingness). This cognitive model of the global view can explain many paralysis phenomena, such as the weakening of purpose due to long-term lack of exercise, or the bias of purpose due to false sensory feedback (hallucinations, false touch). We will verify the plausibility of the model by analyzing previous literature data and clinical observations, such as comparing whether the model prediction is consistent with the changes in brain activity and electromyographic response during the exercise attempt in real paralyzed patients. In addition, the model will also be used to guide the design of subsequent stimulus control algorithms: different DIKWP layers correspond to different control strategy dimensions, such as low-level signal filtering and feedback delay compensation in the data layer, pre-stored motion pattern library matching in the knowledge layer, multi-objective optimization decision-making in the Wisdom layer, and terminal target selection in the Purpose layer. The results of this module will form a cognitive interactive simulation platform, which can verify the hypothetical control strategy in a virtual environment (such as simulating the dynamic changes of each element when the patient tries to walk on the computer), and provide a theoretical basis for subsequent technologies.

Expected results: Complete the customized definition and implementation of DIKWP × DIKWP model in motion control scenarios; Publish relevant theoretical papers to explain the structure and mechanism of the model; Develop a prototype software for cognitive interaction simulation, and verify the effectiveness of the model through examples, such as simulating single-joint motion Purpose Transfer and Feedback Loop.

(2) Artificial consciousness-driven motor purpose training and feedback module

Main content: Develop an artificial consciousness interaction training system to help paralyzed patients reactivate and train the brain's motor purpose generation circuit. The system consists of two sub-parts: the mathematical consciousness subsystem and the physiological consciousness subsystem. Based on the DIKWP model and the "BUG" theory of consciousness, the mathematical subsystem undertakes the functions of identifying, interpreting and guiding the user's purpose information. The physiological subsystem connects with the patient's physiological signals and perceptual feedback, and realizes information interaction with the mathematical subsystem.

Specifically, we'll build a Purpose training scenario on a virtual reality (VR) or augmented reality platform: The patient wears a BCI device (eg, a noninvasive EEG cap) or imaging device and attempts to perform some simple movement (eg, imagining a leg lift, grasping an object) in a virtual environment as prompted. The mathematical consciousness subsystem uses the pre-trained Purpose decoding algorithm to extract the features related to the purpose of movement (data → information conversion) from the patient's EEG/Encephalomagnetic data, and elevates these features to the semantic level through the DIKWP model, and compares them with the existing knowledge base (such as the brain signal pattern of normal people completing the action) (information → knowledge transformation) to judge whether the patient's purpose is clear and complete. If the Purpose signal is detected to be weak or disordered, the artificial consciousness will give feedback through the virtual coach, such as visual/auditory prompts for the correct way to exert force, or even directly replace the patient in the virtual environment to perform the action (the virtual body movement is controlled by an artificial intelligence agent), so that the patient can see that his "Purpose" has been realized, thus forming a Purpose-feedback loopto strengthen the activation of the corresponding pathways in the brain. At the same time, the physiological subsystem records the patient's heart rate, electromyography and other physiological responses, evaluates the patient's concentration and effort (Wisdom layer), and feeds this information back to the mathematical subsystem for adjusting the guidance strategy (e.g., reducing the training intensity when the physiological load is too high, and giving reminders when the attention is decreasing). During the operation of the whole system, the artificial consciousness module acts as a "metacognitive coach" to self-reflect and adjust the patient's purpose attempt: for example, when multiple attempts are unsuccessful, guide the patient to change the imagination strategy; When mastered gradually, the difficulty of the task increases. This kind of interaction based on artificial consciousness is expected to break through the limitations of traditional one-way neurofeedback training and provide a more humanized and effective new mode of rehabilitation training.

Key technical problems: This module needs to solve: (1) motor purpose signal decoding and evaluation: extract the residual motor imagination brain signal characteristics of paralyzed patients, and evaluate the degree of deviation from the normal mode. Machine learning/deep learning methods will be used to construct the Purpose detection model, and the detection results will be interpreted with DIKWP knowledge layer semantics (e.g., "the patient's current purpose is unclear, and there is a suspected lack of expected sensory support"). (2) Artificial consciousness decision-making: design the decision-making mechanism of the consciousness subsystem, that is, how to adjust the training feedback according to the information of each layer of DIKWP. This is equivalent to defining the "state of consciousness" and "behavior" of artificial consciousness, for example, the purpose layer of the consciousness system always aims to "help users successfully produce effective purpose", the knowledge layer masters the rules of various training and guidance strategies, and the wisdom layer selects the best strategy according to the context. (3) Virtual-real feedback mapping: Ensure that feedback in the virtual environment can be effectively transferred to the patient's subjective experience. For example, by providing sensory feedback to the patient on a virtual motion through a synchronized tactile device, or by applying a burst of peripheral nerve electrical stimulation when the patient sees the virtual self moving a leg, the brain receives the leg sensation, thereby tricking the brain into producing a "I moved" experience, so as to enhance the association of purpose with feedback. This involves techniques such as multimodal feedback synchronization and sensory reproduction. (4) System adaptability: the cognitive ability and degree of impairment vary greatly among different patients, and the artificial consciousness system needs to adaptively adjust the parameters, for example, for patients with high paraplegia but good cognition, they can rely more on their active purpose; For cognitive deterioration caused by long-term paralysis, artificial consciousness needs to be more actively involved and guided. We will introduce methods such as reinforcement learning, so that the artificial consciousness system can continuously optimize its own strategy through training data to adapt to individual differences.

Expected Results: This module will result in the development of a set of artificial awareness Purpose training software and supporting user-side hardware (which may include VR glasses, EEG acquisition devices, vibration haptic feedback devices, etc.). Initial testing on a small sample of paralyzed patients is expected to show an increase in brain motor purpose signaling(e.g., increased EEG ERD amplitude), increased success rate of Purpose, and even improved mental state (e.g., increased self-efficacy). We plan to publish academic papers on "Purpose Reconstruction of Artificial Consciousness-Assisted Movement", as well as the formation of software copyrights or patents (such as "Artificial Consciousness Rehabilitation Training Method"). The results of this module will lay the cognitive and purpose foundation for subsequent real-life electrical stimulation training.

(3) DIKWP hierarchical mapping of electrical stimulation precision control module

Main contents: To study and implement a multi-parameter electrical stimulation collaborative control strategy fused with the DIKWP model, the core of which is to organically combine spinal epidural stimulation (SCS) and peripheral nerve/muscle stimulation (PNS/FES) to simulate and restore the central-peripheral physiological signal circulation. We will design the corresponding stimulation modes based on the representation of the target action at different levels in the DIKWP model: (a) Purpose layer: the overall stimulation goal is determined by the patient's current desired functional action, such as standing, walking, or holding, and different stimulation protocols are selected accordingly. (b) Wisdom layer: Adjust the stimulation strategy according to the perception and context of the environment, such as the need to adjust the muscle strength distribution of the legs in real time when the ground slope changes, which can be realized by the Wisdom layer decision engine. (c) Knowledge layer (knowledge): spatiotemporal patterns of spinal-muscular stimulation that store and recall specific movements. For example, walking can be pre-stored with a multi-channel stimulation schedule organized according to gait phase, and a fist can simultaneously stimulate the ulnar nerve and median nerve. The knowledge layer acts as a repository of expert experience, defining the current intensity/frequency combinations that each channel electrode should output at different moments to achieve a specific synergistic action. (d) Information layer (information): deals with low-level sensor feedback and state estimation. For example, electromyography sensing is used to assess the degree of muscle contraction triggered by the current stimulus, and accelerometers are used to determine whether the joint movement is at the desired angle. The information layer compares this data with the expected model of the knowledge layer to generate an error signal. (e) Data layer (data): high-speed processing of raw signals and instructions, including basic operations such as filtering, A/D conversion, and pulse train generation.

According to the above layering, we design the architecture of the multi-parameter control system: the central control algorithm is the implementation of the Wisdom layer and the knowledge layer, running in the host computer or implantable controller; Algorithm inputs include user purpose (possibly from the results of module 2 decoding), environmental information (from sensors, such as pressure pads to sense the force on the plantar foot), and real-time feedback (provided by the information layer). The algorithm first retrieves the initial value of the stimulus mode of the corresponding action in the knowledge base, and then adjusts the mode parameters through the optimization algorithm based on the feedback error to achieve closed-loop correction. For example, when a lack of EMG response is detected, the amplitude of the corresponding channel current can be automatically increased; If joint movement is delayed, it may be necessary to advance the phase of stimulation of the channel. We will explore the method of using model predictive control (MPC) or incremental PID combined with fuzzy control to introduce a certain amount of self-learning ability while ensuring stability. In particular, in terms of dual-stimulation synergy, the division of labor between spinal cord stimulation and muscle stimulation is determined: spinal cord stimulation primarily activates spinal cord networks such as the central pattern generator (CPG) (which provides basic rhythms), and muscle/nerve stimulation is used as a supplement to strengthen specific muscle group contractions (providing additional strength or fine tuning). This is similar to feedforward + feedback control: spinal cord stimulation provides a feedforward drive signal, and muscle stimulation corrects based on actual action feedback. Based on data from previous animal trials and simulations in this project, we will optimize this division of labor, for example by determining the frequency and amplitude ratio of the two during the different phases of exercise. It has been shown that a spinal cord stimulation frequency of 10–20 Hz combined with muscle stimulation can promote pathway reconstruction and functional recovery at the same time. Based on this, we will further refine the optimal frequency/waveform combination for different actions.

Technical difficulties: (1) Multi-objective parameter optimization: Controlling dozens of electrical stimulation channels at the same time to complete multi-joint cooperation is an optimization problem with high dimensions and multiple constraints. We need to take into account the goals of movement accuracy, balance stability, and stimulus safety (to prevent excessive fatigue or tissue damage). Prepare for a hierarchical optimization strategy: first use a heuristic algorithm to determine the overall parameters of the spinal cord and peripheral stimulation (e.g., the ratio of each to the stimulus intensity), and then use a quick search or gradient algorithm to fine-tune the single-channel parameters. (2) Real-time: The control algorithm must run on a millisecond cycle (especially under fast rhythmic movements such as walking, it can be detected with a delay of 100ms). Therefore, the algorithm needs to be efficient and concise, and we will do code optimization and hardware acceleration (such as using FPGA to handle heavy calculations) while maintaining performance. (3) Personalization and self-adaptation: each patient has different neurological residual functions, and the effect of the same stimulation parameters in different people varies greatly. Therefore, the system needs to have an adaptive tuning function. We consider introducing a closed-loop parameter mechanism: the system automatically scans different combinations of parameters and evaluates the output (e.g., testing the effect of different spinal electrode combinations on the EMG response) at the patient's initial use, forming a patient-specific parameter set. Thereafter, in long-term use, if a decrease in the effect of the parameters is detected (e.g. muscle fatigue due to attenuation of response), online parameter adjustment is initiated. The knowledge layer and the wisdom layer of the DIKWP model come into play here: the knowledge layer stores the patient's past valid parameter records, and the wisdom layer uses the rules engine to determine when adjustments are needed and adjust the strategy.

Expected Results: A complete set of spinal-peripheral bistimulation synergistic control algorithms was formed, and its effectiveness was verified in computer simulations and ex vivo/animal experiments. Specific achievements include: the software implementation code base of the algorithm, key technology invention patents (e.g., "Multi-channel neurostimulation control method based on DIKWP model"), and academic papers (e.g., explaining the effect of closed-loop control of double stimulation, such as the improvement of motor function indicators compared with single stimulus). In terms of validation effect, we expect to see that compared with the traditional scheme, the new algorithm can reduce the stimulation current threshold (because the synergistic effect is more effective), improve the accuracy of motion output (reduce the error of joint trajectory), and improve neuroplasticity indicators (such as the enhanced response of the central conduction pathway evoked after stimulation). These will provide a strong basis for the next step of hardware implementation and human trials.

(4) Cognitive perceptual feedback multi-channel stimulation system module

Main contents: Research and development of high-performance neurostimulators with 64 channels and above and their supporting sensing and control modules to achieve collaborative stimulation control of multiple joints and muscle groups, and have cognitive perception feedback functions. The module consists of two parts: hardware and firmware.

On the hardware side, we will expand the number of channels and increase the feedback interface based on the design of the existing neurostimulator. The overall architecture of the system can be divided into: stimulation output unit, signal acquisition unit, main control unit and communication unit. The stimulation output unit adopts a multi-channel constant current source or voltage source design, and supports at least 64 independent programmable outputs, each with an output current range of 020 mA (accuracy < 0.1 mA), pulse width of 01 ms (step adjustable), and a frequency of up to 500 Hz to meet diverse stimulation needs. The output channels can be flexibly assigned to epidural spinal electrode arrays (e.g., 8×8 matrix, 64 points total) or multiple groups of myoelectrodes via a switching matrix. We will design flexible electrode plates for the spinal-epidural space, containing 16 or 32 electrode contacts, arranged in multiple posterior root entry regions corresponding to the lumbosacral segment, to cover the innervated segments of major lower limb muscle groups. In the periphery, a combination of surface electrodes and buried electrodes is used to provide stimulation interfaces for key muscle groups such as quadriceps, gastrocnemius, and tibialis anterior muscles. The signal acquisition unit includes a number of analog front-ends, which are used to collect feedback signals such as electromyography (EMG), joint angle (through wearable IMU), pressure (insole sensor), etc., with a sampling rate of 1~2 kHz and a resolution of 16 bits per channel. The focus is on electromyography feedback, which can help determine the stimulus-induced muscle excitation and contrast voluntary contractions with electrically evoked contractions. The main control unit uses a high-performance embedded processor (such as ARM Cortex-M or RISC-V cores) combined with an FPGA to process massively parallel data. The main control unit executes a closed-loop control algorithm and communicates with the host computer or the implanted consciousness module. The communication unit provides a wireless link (e.g. Bluetooth Low Energy BLE or dedicated RF) for uploading data and receiving high-level instructions. Considering that the system has two parts, implantable and extracorporeal, we will adopt a modular design: the extracorporeal controller is placed on the patient's wheelchair or belt, including the main control and battery; The implanted stimulator is as small as possible and can be placed epidural or subcutaneously to be responsible for the actual electrical stimulation output, and is connected to an extracorporeal controller via percutaneous or wireless means. This split solution reduces the size of the implant and facilitates power delivery.

In terms of firmware and control software, we will develop firmware driven by a real-time operating system (RTOS) to ensure synchronization and low latency of multi-channel stimulation and acquisition. Key tasks include: multi-channel waveform generation, hardware safety monitoring (e.g., impedance detection, temperature monitoring to prevent overheating), data buffering, and packaging and transmission. Cognitive perceptual feedback is reflected in the firmware logic: for example, after each stimulus pulse is emitted, the firmware waits for a preset delay to collect the corresponding EMG response, calculates the contraction amplitude, and decides whether to adjust the intensity of the next pulse compared to the expected value. The firmware needs to provide configurable rules for the high-level DIKWP smart module to issue policies, such as "enable feedback adjustment mode" or "designate a muscle to participate in the closed-loop". We will also develop host computer monitoring software for parameter configuration, data visualization and documentation. The host computer software will integrate the DIKWP structured evaluation module, which can display the indicators of each layer in real time during the training process (e.g., data layer: current myotelem-to-noise ratio; Information layer: current range of motion; Knowledge layer: percentage of action completion; Purpose layer: Patient active cooperation score, etc.), which are valuable for clinicians to assess the effectiveness of training.

Innovations and challenges: (1) At present, there are few neurostimulation systems with more than 32 channels in commercial use at home and abroad, and our 64+ channel system will reach a new height in terms of hardware integration and controllability, and it is necessary to overcome engineering problems such as inter-channel interference and power management. (2) Cognitive perceptual feedback is not available in traditional stimulators, and our system will be the first to combine sensor acquisition and stimulus output in a closed loop on the same platform, so that the stimulator is no longer a "blind" output device, but an intelligent device with preliminary "cognition" - it can "sense" whether the muscles are following the stimulus instructions and adjust themselves accordingly. (3) We will introduce safety protection mechanisms throughout the hardware and firmware design, such as software limiting, preventing continuous single-point high-frequency stimulation, and automatic stopping of abnormal disconnection detection, etc., to ensure that any uncontrolled risk can be contained in time in a closed-loop situation to meet the safety requirements of medical devices.

Expected Results: A set of laboratory prototype devices has been developed, including multi-channel stimulation host, implanted electrodes and accessories, and supporting control software. We will thoroughly test its performance metrics with the following goals: <5% output waveform error, <–60 dB of channel-to-channel crosstalk, < 10 ms control delay, and >8 hours of continuous system operation (battery powered). Its function is then validated in animal experiments (e.g., testing the electromyography response of the hindlimb elicited by multi-point spinal cord stimulation on anesthetized animals, and observing multi-point synergistic gain versus single-point stimulation). If possible, we also plan to conduct preliminary functional tests on several volunteer subjects (patients with complete paraplegia with low thoracic spine who can be recruited and temporarily implanted with epidural electrodes under local anesthesia) to verify that the system can activate the specified muscle groups according to the set program and that the parameters can be adjusted, for example, to change the range of motion of the joints. Success indicators include the induction of coordinated movement of at least 3 lower extremity joints (e.g., standing or pedaling), simultaneous activation of multiple muscle groups and sequential proximity to physiological patterns (as assessed by electromyography timing correlation). The results of this module will be presented in the form of hardware objects and test reports, and relevant invention patents will be applied for (such as "multi-channel neurostimulation device with feedback control") to provide a hardware basis for the final integration.

(5) Integrated intelligent rehabilitation platform integration and clinical application module

Main content: After the research and development of all the above modules is completed, we will carry out system-level integration, build an intelligent rehabilitation platform, and carry out functional verification and clinical testing. The integration work includes software and hardware interface connection, system joint debugging, user interface development and practical application plan formulation.

In terms of software integration, the focus is on seamlessly connecting the artificial consciousness Purpose training module (2) and the stimulus control module (3) and (4) to achieve a complete closed-loop transition from Purpose training to actual movement. When the patient is using it, the platform first runs an artificial consciousness training program, and the patient is trained on several virtual tasks to activate the brain Purpose signal. This is followed by the gradual introduction of real electrical stimulation, allowing virtual movements to occur in parallel with actual body movements, and finally transitioning to fully real motor training. This requires the software to be able to switch control modes according to the training phase, and to ensure that the artificial consciousness feedback and the electrical stimulation control are coordinated without conflict. We will develop a unified scheduling module to manage the operation of each subsystem according to the preset training process, such as the main virtual training in the first week, the addition of electrical stimulation synergy in the second week, and the full physical training in the Nth week. At the same time, the DIKWP structured evaluation module runs throughout, continuously collecting indicators at all levels to generate rehabilitation reports. The platform's software also includes a patient database, which stores each patient's training data, model parameters, and stage evaluation results for personalized adjustment and scientific analysis.

In terms of hardware integration, it is necessary to combine the multi-channel stimulator with the rehabilitation training environment. We will set up a test bed or training room in the rehabilitation institution, equipped with implantable electrodes (if it is an implanted version, the neurosurgeon will need to complete the implantation surgery; In the case of the percutaneous version, the corresponding electrodes are installed), the stimulation unit, the computer, the VR device, etc. Ensure that the communication interface of the stimulator is connected with the host computer software, and the sensor data stream can be transmitted to the cognitive assessment module in real time. Build a safety monitoring system, including ECG monitoring, emergency stop button, etc., to ensure that the stimulation can be disconnected and alarmed immediately in case of abnormal conditions.

Clinical Validation Protocol: We plan to conduct a small-scale clinical trial in a joint rehabilitation hospital or neurosurgery center in the later stage of the project. No less than 5 eligible patients with paralysis (such as complete paraplegia AIS class A or B, level of injury above T12, complete paralysis of the lower limbs for more than 1 year) were recruited to participate in the study with the approval of the ethics committee and full informed consent. The clinical validation will adopt a single-group controlled trial design to evaluate the functional improvement of the platform before and after intervention. The main validation indicators included: muscle strength and motor function level (lower limb muscle strength assessment, Barthel index and other daily living ability scores), evoked range of motion (range of motion), motor purpose signal changes (EEG/EMG ERD power changes), and subjective feelings (pain, fatigue, self-control questionnaires). The training cycle is expected to be 3~6 months, and the training is carried out 3-5 times a week, about 1 hour each time. During the training process, the platform automatically records detailed data. The goal of success was set as: all subjects were able to tolerate the training without serious adverse events; At least 3 patients achieved previously unattainable motor functions after the intervention, such as being able to stand and step with assistance, or achieve partial autonomous control of leg movements in the supine position; The patient's brain motor purpose begins to be correlated with actual movements, for example, after the brain-computer linkage is removed, a patient can walk short distances with crutches with residual voluntary control (ideally, this indicates that the neural pathway is partially reconstructed and autonomic function is restored). In addition, the quality of life questionnaire showed a significant improvement in expected psychological state and self-care confidence. We will also invite rehabilitation doctors to compare and evaluate the differences between this platform and traditional rehabilitation methods, and collect suggestions for improvement.

Promotion and application path: After verifying that the platform is safe and effective, we will actively promote its development in the direction of clinical application and industrialization on a larger scale. Specific measures include: 1) Standardization and adaptive improvement: According to the feedback from the clinical trial, the system is optimized to make it more stable and easy to use, and clinical use specifications and training manuals are compiled for rehabilitation therapists to learn and master. 2) Preparation for regulatory approval: Organize clinical data, prepare application materials in accordance with medical device regulatory requirements, and prepare for the next step of applying for medical device registration. In particular, if implantable electrodes are used, we need to conduct long-term safety tracking and may request a larger clinical trial (e.g., a multicenter trial) to meet regulatory requirements. 3) Industrial cooperation: seek cooperation with medical device companies or investors to productize key technology modules. For example, we work with qualified companies to develop commercial versions of implantable electrodes and stimulators to ensure that medical safety standards and production consistency requirements are met; Artificial consciousness rehabilitation software is packaged as commercial software, and technical support and upgrades are provided. 4) Demonstration application: Strive to establish a demonstration base in the National Center for Neurological Diseases or key rehabilitation hospitals, carry out continuous application, and invite more patients to experience. Through the promotion of typical success stories, the industry and the public will be more aware of the technology. We also plan to incorporate the results of this project into the rehabilitation engineering professional courses and continuing education programs of colleges and universities to cultivate more professionals. 5) Medical insurance and policy support: Actively cooperate with the health authorities and medical insurance departments to evaluate new technologies, provide data on the efficacy and cost benefits of the project, and promote the inclusion of "cerebrospinal interface rehabilitation training" in the price of medical services and the scope of medical insurance reimbursement. This will greatly promote the popularization of this technology.

Through the above path, we will strive to realize the results of this project from the laboratory to clinical application within 5 years after the end of the project, and initially establish an innovation chain combining production, education, research and medicine. It is expected that this project will promote the leapfrog development in the field of intelligent neurorehabilitation in China: on the one hand, it will bring real functional improvement to paraplegic patients, and on the other hand, it will also cultivate a new generation of rehabilitation equipment products with independent intellectual property rights, seize the opportunity of the international market, and create huge social and economic benefits.

Phased research plans and milestones

To ensure that the project objectives are successfully achieved, we have developed a phased research plan with the following milestones:

• Phase 1 (Initial Project ~ End of Year 1): Theoretical Modeling and Key Algorithm Design. He mainly completed research background research, demand analysis, and proof of concept. Specific milestones include: (1) building a preliminary DIKWP ×DIKWP cognitive interaction model framework, completing the theoretical analysis of the closed-loop mechanism of motor purpose and feedback, and writing a model definition and hypothesis report; (2) develop a basic motion purpose recognition algorithm and artificial consciousness decision-making logic, and verify the feasibility of artificial consciousness guiding motion imagination in a simulation environment. (3) Design a prototype of multi-parameter electrical stimulation control algorithm, and use MATLAB/Simulink and other tools to establish a dual-stimulation control simulation model to simulate the control effect of simple movements (such as knee flexion and extension); (4) Demonstration of hardware scheme, complete the overall scheme design and selection of key components of the multi-channel stimulator, and proofing the single-channel/few-channel function board to verify the output accuracy and safety. Milestone checkpoints: Submit the Requirement Analysis and Schematic Design Report; Publish/submit at least 1 academic paper on theoretical methods (application of DIKWP cognitive model in motion control); Completed the development demonstration of the prototype software for artificial consciousness training.

• The second stage (2nd year ~ 3rd year): core system development and initial integration. Focus on hardware development and algorithm implementation, and implement each module into an actual system. Specific milestones: (1) Completed the development of 64-channel stimulator hardware, including circuit board design and manufacturing, embedded firmware writing, and achieved performance indicators through laboratory testing; (2) Improve the function of the artificial consciousness training system, integrate the virtual environment, EEG acquisition, and haptic feedback equipment with the software, carry out the trial experience of healthy subjects, and adjust the algorithm parameters to improve the robustness; (3) realize the transplantation and optimization of the closed-loop control algorithm under the guidance of DIKWP on the embedded platform, so that it can run in real time on the main control of the stimulator; (4) Functional validation experiments were carried out on small animal models, such as testing the effect of double stimulation on hind limb motor function indexes in rats with partial spinal cord injury (in cooperation with animal experimental institutions), and collecting neural remodeling biomarker data; (5) Preliminary integration of artificial consciousness module and stimulus control module, end-to-end test on simulated human environment (such as simulated human model or lower limb robot) to verify the process coherence and correctness from Purpose input to stimulus output. Milestone checkpoints: submission of interim research reports and presentations of milestones; Apply for at least 2 invention patents (one for hardware systems and one for control algorithms); He has published 2 papers in international conferences/journals (e.g., on the development of dual-cycle artificial consciousness algorithm and multi-channel stimulation device). The mid-term evaluation focuses on checking whether the hardware equipment meets the design requirements, whether the algorithm performance meets the real-time closed-loop requirements, and whether the animal experiments prove that the method is effective.

• Phase 3 (Year 4): System Integration and Security Assessment. Fully integrate all subsystems in the laboratory to carry out rigorous safety and reliability testing and simulated use evaluation. Milestones: (1) Completed the software integration of the intelligent rehabilitation platform, including a unified user interface and data management functions, and realized the one-click switching between artificial consciousness training and stimulation control; (2) Formulate system safety specifications, complete risk analysis and redundant protection measures, and realize the automatic detection and emergency disposal process of unexpected situations (such as sensor failure and patient physiological abnormalities); (3) Conduct long-term continuous operation test (at least 72 hours of uninterrupted operation) on the integrated system to evaluate the system stability, temperature rise, data accuracy, and improve heat dissipation and power management; (4) Carry out functional verification experiments: recruit a number of able-bodied volunteers to simulate wearing equipment for some training tasks to test the effectiveness and comfort of the system in artificial simulation situations (for example, let able-bodied people try to control the FES of the arm through EEG to cause finger grasping action to verify the closed-loop performance of the system); (5) Organize reviews with project consultants and experts (in the fields of rehabilitation medicine and neural engineering) to propose improvement plans for the ease of use and potential problems of the system. Milestone checkpoints: safety and reliability test reports; Prototype of a fully integrated system; Through the internal acceptance of the expert group, the approval opinion for entering the clinical trial was obtained. Whether there is a clinical trial condition is a key evaluation indicator at this stage.

• Phase 4 (Year 5): Clinical trials and application demonstrations. After obtaining the approval of ethics and relevant departments, the clinical trial and effect evaluation of paralysis patients will be officially launched. Milestones: (1) Completed the first patient implantation surgery and system installation, carried out a detailed training plan, continuously adjusted the parameters during the training, and recorded the whole process data; (2) Expand the trial to multiple patients, collect effect data of different injuries and different populations, and enrich the diversity of samples; (3) Comparative analysis of clinical indicators: such as the improvement of lower limb muscle strength/control scores before and after training, changes in patients' EEG Purpose signals, improvement of daily function, etc., and extracting statistical results; (4) Organize all test data and cases, convene expert seminars to evaluate the effectiveness, safety and improvement space of the technology, and formulate guidance for further promotion; (5) Prepare draft clinical application guidelines, including patient selection criteria, operating procedures, precautions, etc., in preparation for wider promotion. Milestone checkpoints: complete the training of no less than 5 patients to obtain a complete clinical data set; Write the "Clinical Trial Report"; Publish papers in authoritative academic journals (the goal is to publish more than 2 SCI papers to explain the innovative efficacy of this project in clinical practice); The technical achievements have passed the project acceptance organized by the competent department.

Among the above-mentioned stage arrangements, the 13th year is the technical research and development stage, and the 45th year focuses on clinical verification and improvement. We will organize a stage review at the end of each stage, and adjust the focus of follow-up work in a timely manner according to the review opinions. In particular, for the clinical stage, we will determine the speed of advancement based on the performance of the system in the early stage to fully ensure the safety of the subjects. The overall project schedule is as follows:

• Year 1: Theoretical Model Construction (Completed); Purpose Decoding and Artificial Consciousness Prototype (Done); Stimulation algorithm simulation (completed); Hardware scheme design (completed).

• Year 2: Development of multi-channel stimulation hardware (prototype completed); Improvement of the artificial consciousness system (completed); control algorithm embedding implementation (completed); Animal experiments are started (implemented).

• Year 3: Animal experiments completed (completed); Initial integration of the system (completed); safety testing (carried out); Application for clinical approval (submission).

• Year 4: Clinical trial initiation (1-2 cases); System optimization and improvement (continuous); More patient recruitment (conducted).

• Year 5: Clinical trial completion (≥ 5 case data); Data analysis and summarization (completed); Release and acceptance of results (completed).

Through the above milestone management, we will identify problems in a timely manner, adjust the plan, and ensure the final successful completion of the project's intended goals.

Feasibility and research basis

This project integrates the multidisciplinary frontiers of cognitive science, artificial intelligence, neural engineering and rehabilitation medicine, although it is very challenging, but it is highly feasible based on the solid research foundation and conditions of our team:

1. The project team has outstanding advantages and a strong academic foundation: Professor Duan Yucong, the project leader, is an internationally renowned expert in artificial intelligence and cognitive computing, and has made a series of leading achievements in the field of DIKWP model and artificial consciousness. The DIKWP mesh cognitive model proposed by Professor Duan realizes multi-directional semantic feedback through mesh interaction, and the related research has been authorized by 114 domestic and foreign invention patents, providing a new path for AI interpretability and artificial consciousness construction. These patented technologies form a complete DIKWP technology system, including cognitive models, semantic operating systems, artificial consciousness architecture and other aspects. For example, Professor Duan's team has developed the world's first small-model, low-computing power explainable artificial consciousness software system DIKWP-AC, which is divided into two subsystems, mathematical and physiological, which can realize the interpretation and interaction of users' physiological and mathematical data on low-computing devices. The system has been initially applied in the field of medical diagnosis and has achieved remarkable results. These original results show that the team has a deep accumulation in artificial consciousness model design, semantic layer computing and cognitive interaction, which can be directly used in this project.

At the same time, the project also has a strong cooperative team and foundation in the field of neuromodulation and rehabilitation. The core members of the project team include experts in neurosurgery and rehabilitation medicine, who have presided over relevant scientific research projects and carried out clinical research on spinal cord stimulation, and have accumulated valuable experience. For example, the team of a rehabilitation hospital participated in the national project of "Brain-Computer Interface Reconstruction of Motor Function", and conducted research on the brain signal control exoskeleton of paraplegic patients, which has rich patient recruitment and evaluation capabilities. The team also maintains cooperation with domestic enterprises engaged in the research and development of implantable medical devices, and can obtain hardware support such as epidural electrodes and implanted signal amplifiers. All these provide a practical basis for the project in the development and clinical implementation of neural interface hardware.

2. The progress of the preliminary work supports the key technologies of the project: Before the project application, we have carried out a series of pre-research work, and the preliminary results obtained prove the feasibility of the technical route of the project:

• In terms of cognitive model, we have completed an application exploration of the DIKWP model in the medical field, which constructs a doctor-patient dialogue artificial consciousness simulation system, maps the information flow in the process of consultation to the five-layer cognitive process of DIKWP, and simulates the prototype of artificial consciousness in doctor-patient interaction through the state machine. This work gives us an intuitive understanding of how the DIKWP model interfaces with physiological processes. For example, we have clarified in the model that the data layer corresponds to physiological signals, the knowledge layer corresponds to the medical knowledge base, and the Purpose layer corresponds to the purpose of doctor-patient communication. This laid the foundation for the use of DIKWP for motion control in this project.

• In terms of artificial consciousness algorithms, the team put forward the famous consciousness "BUG" theory, which compares the human consciousness process to a continuous "text solitaire" process, where most of the information processing is completed automatically in the subconscious, and only when an "error" occurs, it arouses conscious attention. Based on this, we developed an anomaly detection and arousal mechanism for artificial consciousness decision-making, which will be used in motor purpose training to determine when artificial consciousness should intervene for adjustment. FOR EXAMPLE, WHEN A PATIENT TRIES AN ACTION SEVERAL TIMES WITHOUT SUCCESS, THE SYSTEM DETECTS THE ACCUMULATION OF DEVIATIONS (THE APPEARANCE OF "BUGS"), WHICH TRIGGERS ARTIFICIAL CONSCIOUSNESS TO PROVIDE A NEW GUIDANCE STRATEGY. This mechanism has been proven in the medical dialogue system we have developed, and it is very effective in improving the efficiency of interaction.

• In terms of electrical stimulation control, the engineers of the project team have participated in the development of a 16-channel electromyographic stimulation rehabilitation instrument for upper limb motor function training of stroke patients, which has passed the prototype test. In the process, we have mastered the basic methods of multi-channel electrical stimulation synchronous output and electromyography feedback acquisition, and developed the corresponding software interface. This is similar to the stimulator design of this project, which can reuse related software and hardware modules. In addition, we have built a simple human lower limb biomechanical model and FES simulation platform, simulated the knee angle control in Matlab, and achieved tracking control of flexion and extension movements by stimulating the quadriceps and hamstrings through dual channels. The simulation results show that the closed-loop control can significantly reduce the angular error, which is better than the open-loop stimulation. This provides us with the confidence to expand to multi-joint multi-channel control.

• In terms of clinical resources, our team worked closely with the Rehabilitation Department of Hainan Provincial People's Hospital to jointly establish the "Intelligent Rehabilitation Joint Laboratory". The hospital has agreed to open up the source of the disease for this project and assign doctors with rich experience in SCI rehabilitation to participate in the project. We have conducted initial follow-up and screening of several patients with spinal cord injury, collecting their basic information and needs. This means that once the project is in the clinical phase, we can expect to start the trial quickly without having to establish a clinical connection from scratch.

3. Perfect scientific research conditions and guarantees: The supporting units provide sufficient scientific research conditions for the project. We have provincial and ministerial key laboratories, equipped with first-class software and hardware facilities in the direction of cognitive computing and brain-like intelligence. The lab has a high-performance computing server farm for deep learning model training and large-scale simulation; Equipped with EEG EEG acquisition equipment, EMG signal analyzer, high-precision 3D motion capture system, etc., it can meet the needs of physiological signal acquisition and kinematic analysis. In addition, the electronic process center of the relying unit can support multi-layer circuit board manufacturing and system integration testing, and provides the necessary instruments for the development of multi-channel stimulators (such as high-speed oscilloscopes, signal generators, EMC test equipment). During the clinical trial phase, the partner hospital will provide beds, operating rooms and rehabilitation training venues to ensure that the trial is conducted safely. The project budget has also fully considered the costs of reagents, animal experiments, clinical subsidies and other expenses, and the funds are guaranteed.

4. Risk response measures: We fully anticipated the technical challenges of the project and formulated corresponding plans. For example, if the implantable electrode approval process is affecting the schedule, we have prepared an alternative: a clinical trial with a percutaneous electrode to verify the results before applying for implantation for long-term use. In terms of artificial consciousness algorithms, if the initial effect of the whole model is not ideal, we can also settle for the next best thing, adopt a more mature training mode based on EEG feedback, and gradually integrate cognitive elements to reduce the difficulty of implementation. At the same time, our multidisciplinary team helps to control risks holistically: engineering staff are responsible for technical details, and medical staff strictly monitor safety and ethics to ensure that the project is on track.

In summary, we have an excellent team, solid preliminary results and complete scientific research conditions, and we are full of confidence in the completion of this project. Each key technology of the project has a corresponding foundation and reserves, and the combination of innovative theoretical frameworks (DIKWP× DIKWP and artificial consciousness) and specific implementation means has been verified and feasible on a small scale. With the development of the project, we will continue to learn from the latest research progress at home and abroad (such as the relevant results published in Nature and other journals) to maintain the advanced nature and correct direction of the program. With the strong support of feasibility and foundation, the smooth implementation of the project is fully guaranteed.

Expected Outcomes and Appraisal Indicators

The project is expected to produce a wealth of scientific research and application results, and we will set assessment indicators from the two levels of scientific research and technology application to comprehensively evaluate the effectiveness of the project:

(1) Scientific research achievements:

• Monographs: Published no less than 8 papers in high-level academic journals or conferences at home and abroad. Among them, there are no less than 5 SCI journal papers (strive to be published in IEEE TBME, Neurorehabilitation and Neural Repair, Nature and other high-impact journals), and no less than 3 Chinese core journal papers. These papers will cover the theory of DIKWP cognitive model, artificial consciousness rehabilitation methods, neurostimulation control techniques, and clinical trial results. At the end of the project, write 1 English monograph or chapter, systematically summarize the theoretical and practical experience of the project.

• Invention patents: apply for no less than 6 Chinese invention patents, and strive to authorize more than 4 patents; At the same time, apply for 1-2 PCT international patents as appropriate to enhance international influence. The patent intends to cover: core innovations such as DIKWP-based rehabilitation training methods, artificial consciousness interaction systems, dual-feedback neurostimulation control devices, and multi-channel stimulator circuits.

• Standard specification: Draft more than 1 group or industry standard. For example, the "Technical Specification for Multichannel Neural Electrical Stimulation Rehabilitation System" or the "Technical Guide for Artificial Intelligence-assisted Sports Rehabilitation Training" were formulated to lay the foundation for subsequent standardization in related fields.

• Academic impact: The core results of the project will be displayed at the International Conference on Artificial Intelligence or Rehabilitation (at least 3 invited reports) to improve China's voice in the intersection of artificial awareness and intelligent rehabilitation. We will host one domestic academic symposium to promote the project concept and cultivate follow-up research capabilities.

Assessment indicators: 8 papers≥ (5 SCI≥), 6 patent ≥applications (4 authorized ≥), and 1 standard/guideline ≥.

(2) Achievements in technical application:

• Prototype system: Completed the development of 1 prototype of intelligent spinal-peripheral nerve interface rehabilitation system, including: 64-channel electrical stimulator host and supporting electrodes, power supply and safety accessories, artificial consciousness training and control software platform, user interface and data management system. The system functions as designed and operates stably in both simulation and clinical environments. Submit a complete physical system and user manual at the time of project acceptance. *Assessment indicators: *The prototype system has multi-joint collaborative control and closed-loop feedback functions, and has been tested to meet the performance indicators (see hardware performance indicators above).

• Clinical efficacy: Achieve the desired functional improvement in the trial patients. At least 3 patients with AIS grade A/B paraplegia regained the ability to make their own initiatives, such as standing and walking ≥ 10 meters with assistance, or achieving active control of paralyzed limbs (even partial movements, such as hip flexion) through this intervention; Other patients also showed varying degrees of improvement (muscle strength level improvement ≥1 grade, daily function assessment score improvement ≥10 points). Objective data collected should show significant improvements, such as an increase in the amplitude of motor evoked potentials after training, and a significant increase in EEG signal power. *Assessment indicators: *The number of clinical cases was ≥5, the functional improvement rate was ≥60%, and no serious adverse events were reported.

• Platform integration and ease of use: Developed an integrated rehabilitation platform software, which can be operated and used by rehabilitation therapists in a clinical environment. The operation process of the system is simple and the safety is high. A questionnaire survey was conducted on the medical staff participating in the trial, and the satisfaction rate was not less than 80%. Patient satisfaction (subjective rating of the training process and effect) is not less than 80%. *Assessment indicators: * The platform software has complete functions, and the user satisfaction ≥ 80%.

• Data and database: Establish a motor purpose-stimulation-feedback dataset for paralyzed patients, including multimodal data (EEG, EMG, exercise, stimulation parameters, etc.) of at least 5 patients trained throughout the whole process, with a data length of more than 100 hours and a total amount of > 1TB. The database will provide a valuable resource for subsequent in-depth analysis and algorithm improvement. Open part of the anonymized data to the scientific research community as appropriate to enhance the impact of the project.

• Talent training: Through the implementation of the project, no less than 5 doctoral/master's students and 2 postdoctoral fellows will be trained to form an interdisciplinary team. Some of the outstanding members will stay in universities and hospitals to continue to promote research or application in this direction and realize the inheritance of intellectual achievements. During the project, there will also be cross-training of engineering and medical personnel, such as organizing training courses for doctors to learn the operation of new equipment, and organizing engineers to practice in the rehabilitation department, so as to deepen the integration between them.

Assessment indicators: 1 set of prototype system; The functional improvement rate of patients was ≥60%; User satisfaction ≥ 80%; 1 dataset was formed; There are 5 graduate students ≥.

(3) Socio-economic benefits:

Although the project is mainly R&D in nature, we have also evaluated the potential socio-economic benefits as an additional indicator: it is expected that within 5 years after the end of the project, if the technology is successfully transferred, it can serve thousands of spinal cord injury patients, and each patient with severe paraplegia can save about 50,000 yuan per year in nursing expenses, which will have huge social benefits. At the same time, the output value of related industries is expected to reach 100 million yuan. We'll track subsequent conversions of your project and include them in your long-term assessment.

In short, the achievement assessment of this project will combine quantitative indicators with qualitative evaluation, focusing not only on the quantity and quality of hard results, but also on the verification of actual clinical effects and feedback on user satisfaction. We will carry out mid-term and final self-evaluation in strict accordance with the above-mentioned indicator system to ensure that all expected goals are completed on schedule and with high quality.

Clinical Validation Plan and Achievement Promotion Pathway

1. Clinical Validation Program: The ultimate goal of this project is to apply to patients with paralysis, so clinical validation is a crucial part. We will steadily advance clinical trials in accordance with medical device and clinical research specifications to ensure scientific, reliable and meaningful conclusions.

• Subject selection: According to the previous protocol, we plan to recruit 510 patients with complete spinal cord injury (ASIA grade A or B), with a predominantly thoracolumbar injury level (avoiding high neck injury to reduce the risk of the trial), age 2050 years, and more than 1 year post-injury time (entering the chronic stable phase). The inclusion criteria include: stable physical and mental condition, no serious complications, able to cooperate with training, and strong willingness to recover; Exclusion criteria include: risk of deep infection, severe osteoporosis, pacemaker and other conditions that are not suitable for electrical stimulation.

• Ethical approval and informed consent: Before initiating clinical trials, we will submit research protocols to the medical ethics committees of partner hospitals and supporting units to ensure that they meet the ethical requirements of human trials. All subjects were required to sign an informed consent form before participating to clearly understand the purpose, process and possible risks of the trial.

• Trial group design: all subjects received the intervention of this project, and there was no no intervention control group (considering that the complete control was inhumane and the significance of the comparison was limited due to the small sample size). However, we will collect a series of baseline data from patients before the intervention as a control reference. In addition, indirect comparisons with historical data from previous conventional rehabilitation therapies may be considered.

• Intervention process: After the subjects are enrolled, the implantation surgery (if the implanted electrode regimen is used) is performed first. After the postoperative recovery period (about 2 weeks), the interventional training will begin. Training is usually scheduled in an outpatient/day rehabilitation center for 60-90 minutes 3-5 times a week, and the entire trial intervention period is approximately 3-6 months. The training content includes: artificial consciousness-guided motor purpose imagination exercises, electrical stimulation-induced action exercises with increasing difficulty, and comprehensive task-oriented training (such as standing transfer, walking and other situational simulations). Assisted by rehabilitation therapists on site, technicians monitor the operation of the equipment. During the training process, special attention is paid to the gradual establishment of patient active participation: the system may be automatically led in the early stage, and the patient's voluntary triggering action may be fought for in the later stage.

• Safety monitoring: The subject's vital signs are monitored throughout the process, focusing on skin condition (to avoid burns or pressure damage in the electrode area), spasticity (to prevent fractures caused by inducing severe muscle spasms, etc.), and autonomic reactions(high paraplegic anti-autonomic hyperreflexia) and so on. In case of abnormalities such as sudden increase in blood pressure, excessive sweating, persistent increase in heart rate, etc., immediately suspend stimulation and carry out medical intervention. Check the electrode position and wound healing regularly. We have detailed adverse event contingency plans, equipped with first-aid medications and equipment to respond to possible medical emergencies.

• Efficacy assessment: Subjects were assessed in multiple dimensions before the intervention (Baseline), during the interim (e.g., 8 weeks), and at the end of the intervention, including:

  • Neurological function assessment: the changes in sensory and motor grades were recorded using the international ASIA rating standard scale. Lower Extremity Motor Score (LEMS); Assessment of muscle tone (Ashworth scale).

  • Functional independence: Improvement in functional independence of living was assessed using the Functional Independence Measure of Spinal Cord Injury (SCIM) or the Barthel index.

  • Athletic ability: Specific movement tests, such as standing time in parallel bars, stride distance, six-minute wheelchair pushing test distance, etc., may be adjusted according to the actual situation of the patient.

  • Objective indicators: evoked electromyography (MEP) assessment, recording the changes in the threshold and amplitude of peripheral electromyography response under electrical stimulation or brain stimulation; Gait analysis (if you can walk, record the change in stride length and gait speed); Balance ability (balance plate test).

  • Purpose signaling: If possible, some patients are given EEG/EMG laboratory tests before and after training, such as trying to imagine movements to see if the ERD/ERS changes increase.

  • Psychosocial: Questionnaires were used to investigate depression and anxiety scales, quality of life (e.g., SF-36), and to understand the impact of rehabilitation on psychological state.

The results of each patient's assessment will be archived and, if necessary, we will ask independent experts to conduct blind reviews of the footage and data to improve the objectivity of the conclusions.

• Statistical analysis: Due to the limited sample size, we mainly used descriptive statistics to perform significance tests (such as paired t-tests or non-parametric tests) based on the before and after changes of each case of their own controls. If the sample is large enough, the average improvement rate is also calculated and confidence intervals are given. It is important to report the details of each patient, as small trials are more focused on individual outcomes and feasibility statements. We also plan to conduct an exploratory analysis of differences in response between patients with different levels of injury and different age lines to provide clues for future improvements.

2. Promotion and application path: After completing the clinical verification and proving the safety and effectiveness of the system, we will start to promote the diffusion and application of the results, and the main steps are as follows:

• Registration and approval: Organize clinical data and apply for a medical device registration certificate for this product according to the approval process of the State Food and Drug Administration for innovative medical devices. If it is not possible to register directly at the end of the project, we will also actively seek to join the special approval channel for innovative medical devices to accelerate the transformation. We prepare sufficient technical documentation, including product specifications, manufacturing processes, clinical data reports, etc., to ensure that regulatory requirements are met.

• Industry incubation: The relying unit or partner has a plan to set up a technology-based start-up company to promote the industrialization of the project's products. The engineering staff and management personnel in the project team will participate in the formation of the company, and strive to complete the company registration and initial financing within one year of the completion of the project. Then the company takes the main body to obtain the production license and establish the production line. We have assessed the preliminary market demand, and even if only 10% of the tens of thousands of new paraplegic cases and hundreds of thousands of existing patients in China use this system every year, the market size is considerable. Startups can quickly occupy the market by cooperating with rehabilitation equipment manufacturers or licensing.

• Clinical demonstration and training: 2-3 large-scale rehabilitation centers are selected to carry out clinical demonstration applications. Through the successful cases of the demonstration base, more medical peers are invited to observe and learn. The team will develop detailed training materials and conduct workshops to teach doctors and therapists in other hospitals how to use the system, including patient screening, electrode implantation techniques, training program development, etc. We aim to train more than 50 professionals who can operate the system within three years and form an application network.

• Inclusion in the rehabilitation system: Communicate with the health authorities to promote the inclusion of "central-peripheral nerve interface training" in the routine rehabilitation treatment. For example, it has been included in the updated version of the Chinese Guidelines for Spinal Cord Injury Rehabilitation, as one of the recommended therapies, to increase its acceptance in the industry. It is also possible to issue expert consensus or guidelines with relevant societies to clarify the indications, operating procedures, evaluation methods, etc., and lower the threshold for each unit to carry out this technology.

• Continuous improvement and upgrading: The feedback collected during the promotion process will feed back into R&D. We plan to establish a user feedback mechanism, and the company arranges technical support personnel to regularly visit various application units to record problems and provide solutions. At the same time, we will pre-develop next-generation products, such as a fully implantable system that more closely integrates brain signal acquisition and stimulation, or customized versions for other indications such as upper limb paralysis and stroke, to maintain technological leadership. In the future, it can also be considered to connect with the national key project of brain-computer interface, and expand this product with the help of national strength.

• International cooperation and output: The technology of this project is at the forefront of the world, and we will actively participate in international exchanges, promote it in academic conferences, and publish the results in foreign journals to enhance its international influence. Once the domestic application certificate is mature, you can seek overseas certification and open up the international market. In particular, developing countries along the "Belt and Road" have a huge demand for low-cost and efficient rehabilitation technology, and we can export this achievement through technology transfer and cooperative medical services to serve more patients.

Overall, after the project is completed, our work will not stop, but will enter a new phase of conversion promotion. Through the combination of scientific research and industry, we can finally make innovative achievements truly applied to clinical practice and benefit patients, which is our original intention. It is believed that with the strong support of national policies and our effective promotion measures, the "spinal-peripheral nerve interface regulation and rehabilitation system based on DIKWP×DIKWP model and artificial consciousness" will not only stop at the research results, but also become an advanced rehabilitation tool for daily use in medical institutions, and promote traditional rehabilitation medicine to move towards a new era of intelligence and personalization. We expect that the successful implementation and promotion of this project will bring new hope to paraplegic patients, win an international reputation for China in the field of brain-computer interface and intelligent rehabilitation, and produce significant social and economic benefits.