A method and system for generating an assisted protocol for dysphagia rehabilitation

Abstract

The application discloses a kind of swallowing disorder rehabilitation auxiliary scheme generation method and system, belong to medical rehabilitation intelligent evaluation technical field, to solve the problem of swallowing training path optimization difference;The present application collects myoelectricity, tongue pressure, laryngeal image and acoustic signal in the process of swallowing by multi-channel sensing device, constructs unified multi-modal swallowing behavior feature vector, realizes obstacle stage recognition and function grade discrimination in combination with pre-training model and historical data, and outputs structured evaluation result;Based on the training action of high matching degree of knowledge graph matching, the personalized training path is generated by fusing muscle group matching degree, edge weight and function grade factor, and the intensity, frequency and rhythm parameters are set;Real-time monitoring action and myoelectricity deviation in training process, output prompt signal and record behavior completion degree;According to feedback, dynamically update rehabilitation progress graph, if deviation is out of limit, then automatically reconstruct path and synchronize doctor intervention, thereby improve rehabilitation efficiency and safety.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of medical rehabilitation intelligent evaluation, more specifically, the present application relates to an auxiliary scheme generation method and system for swallowing disorder rehabilitation. BACKGROUND

[0002] Swallowing disorder is a common functional disorder problem in patients with Parkinson's disease, postoperative head and neck, and senile degenerative diseases, etc. Its rehabilitation process usually relies on professional evaluation personnel to judge the disorder degree through observation, scale scoring and other means, and to formulate rehabilitation training plan based on clinical experience. The existing rehabilitation method mostly uses standardized action templates for training, such as forced swallowing, tongue propulsion, glottis closure, etc.

[0003] The existing technology has the following deficiencies: the existing swallowing disorder rehabilitation method generally relies on static templates or experience rules to match training actions in the training path development process, lacks structured understanding and calculation expression of real-time evaluation results, functional grade labels and muscle group restriction information of patients, and is difficult to realize training action optimization and parameter configuration based on multi-dimensional adaptive factors, resulting in lack of personalized adaptability of training path. Especially when facing patients with different disorder sites, different severity and compliance changes, the current method cannot dynamically adjust the training action sequence or reconstruct the action combination, lacks precise matching mechanism of training intensity parameters, frequency parameters and rhythm parameters, thereby affecting the targeting and continuous effectiveness of the rehabilitation process, and limiting the intelligent and fine level of swallowing rehabilitation. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides an auxiliary scheme generation method and system for swallowing disorder rehabilitation to solve the problem of poor swallowing training path optimization in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] An auxiliary scheme generation method for swallowing disorder rehabilitation, comprising the following steps:

[0007] Based on the multi-channel sensing device, the myoelectric signal, tongue pressure data, laryngeal movement image and speech feature in the swallowing process are collected, the multi-modal swallowing behavior feature vector is constructed, and is uniformly mapped into a standard time frame sequence;

[0008] Combined with the pre-trained swallowing function evaluation model and historical rehabilitation data, the disorder stage recognition and function grade discrimination are performed, and the structured evaluation results containing disorder type, site and severity label are outputted;

[0009] According to the evaluation result, relevant training action nodes in the rehabilitation knowledge graph are matched, the adaptation degree weight of the training action is calculated, and an action ranking list is generated, thereby constructing a personalized training path containing intensity, frequency and rhythm parameters;

[0010] During the training execution process, interactive monitoring is performed synchronously, and according to real-time action capture and electromyographic response recognition algorithms, training deviations are recognized and prompt signals are output, and behavior completion degree and deviation level indicators are recorded;

[0011] According to the training feedback record and the stage score, the rehabilitation progress graph is updated, and if the deviation level is continuously identified to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronized to assist in strategy optimization and intervention.

[0012] In a preferred embodiment, based on a multi-channel sensing device, electromyographic signals, tongue pressure data, laryngeal movement images and voice features during swallowing are collected, a multi-modal swallowing behavior feature vector is constructed, and is uniformly mapped into a standard time frame sequence, the specific process is as follows:

[0013] Through a surface electromyography electrode array, time sequence electromyographic voltage signals generated by muscle groups located in the mandible, larynx and sublingual area during swallowing are obtained and recorded as electromyographic signal data;

[0014] Through a tongue pressure sensing pad, unit pressure change values of the tongue and the contact area of the upper palate are obtained, and the pressure duration is recorded synchronously, which is recorded as tongue pressure data;

[0015] An infrared vision camera is used to obtain the vertical lifting trajectory of the laryngeal prominence during swallowing, and the lifting speed and amplitude are calculated, which are recorded as laryngeal movement image data;

[0016] A close-range pickup microphone is used to collect acoustic response signals during swallowing, and the starting time, peak frequency and duration are extracted, which are recorded as voice feature data;

[0017] The four types of heterogeneous sensing data, electromyographic signal data, tongue pressure data, laryngeal movement image data and voice feature data, are uniformly mapped into a standardized time frame structure with the swallowing event start as the reference point, forming a swallowing behavior feature vector set.

[0018] In a preferred embodiment, combined with a pre-trained swallowing function evaluation model and historical rehabilitation data, disorder stage recognition and function level discrimination are performed, and structured evaluation results containing disorder type, location and severity labels are output, the specific process is as follows:

[0019] The constructed swallowing behavior feature vector is input into the pre-trained swallowing function evaluation model, and the swallowing function evaluation model includes a convolution structure for extracting static features and a recurrent neural network structure for modeling time sequence changes;

[0020] The response capability of the key frame segment of the swallowing function evaluation model is enhanced using an attention mechanism to determine the feature differences in the swallowing initiation period, laryngeal elevation period, and contraction recovery period;

[0021] After the swallowing function evaluation model processes the input swallowing behavior feature vector, a set of structured prediction results is output, including the classification probability prediction of the disorder stage, and the disorder stage involved in the swallowing process is identified in the output structured prediction results, including the oral preparation period, the pharyngeal period, and the esophageal transport period, and the corresponding disorder type label is output;

[0022] The swallowing function evaluation model prediction result is matched with the sample with a clear functional level label in the historical rehabilitation database, and the current functional level range is confirmed by a similarity measurement method;

[0023] The structured evaluation results are generated by combining the swallowing function evaluation model recognition results and the historical score mapping relationship, and the evaluation results include the disorder type, the main restricted site, and the severity level label.

[0024] In a preferred embodiment, the swallowing function evaluation model prediction result is matched with the sample with a clear functional level label in the historical rehabilitation database, and the current functional level range is confirmed by a similarity measurement method, and the specific process is as follows:

[0025] A standard sample set containing functional level labels is extracted from the historical rehabilitation database, and each sample in the standard sample set includes a swallowing feature vector of the same modal dimension as the current patient and its corresponding verified functional level label;

[0026] Feature normalization processing is performed on the feature vector output by the current swallowing function evaluation model;

[0027] The feature distance between the normalized current feature vector and each sample in the standard sample set is calculated in turn, and the feature distance calculation is based on the cumulative results of the numerical value differences between the modes, including the electromyographic response difference, the tongue pressure peak difference, the laryngeal trajectory change difference, and the acoustic duration difference;

[0028] According to all distance calculation results, a sample similarity sorting list is generated, and at least two samples with a similarity higher than a set threshold are selected as reference samples from the list;

[0029] The corresponding functional level label distribution in the reference sample is counted, and the functional level range of the current evaluation object is determined according to the highest frequency level label, which is the final output severity discrimination result.

[0030] In a preferred embodiment, relevant training action nodes in the rehabilitation knowledge graph are matched according to the evaluation results, the fitness weight of the training actions is calculated, and a ranking list of actions is generated. The specific process is as follows:

[0031] Based on the structured assessment results, the types of obstacles, restricted areas, and severity levels are extracted and used as a set of nodes to be matched, which are then input into a pre-constructed rehabilitation knowledge graph.

[0032] The rehabilitation knowledge graph is a heterogeneous graph structure, which contains the association edges between training action nodes and obstacle type nodes. The edge weights represent the degree of adaptation of the action to a specific obstacle.

[0033] Starting from the obstacle type node in the rehabilitation knowledge graph, a breadth-first search is performed to traverse adjacent action nodes and extract all directly or indirectly connected candidate training action nodes.

[0034] For each candidate training action node, the fit weight between the candidate training action node and the evaluation result is calculated. The fit weight is determined by the edge weight, the matching degree of the muscle group targeted by the action, and the fit factor of the patient's current functional level.

[0035] All candidate training action nodes are sorted from high to low according to their fitness weight to generate an action sorting list, which serves as the input candidate set for constructing the training path.

[0036] In a preferred embodiment, for each candidate training action node, the fit weight between the candidate training action node and the evaluation result is calculated. The fit weight is jointly determined by the edge weight, the matching degree of the muscle group targeted by the action, and the fit factor of the patient's current functional level. The specific process is as follows:

[0037] Obtain information on the main restricted muscle groups in the evaluation results, compare them with the target muscle groups involved in the candidate training action nodes, and calculate the muscle group matching index. The muscle group matching index is quantified based on the overlap of the target muscle groups. Complete coverage is recorded as 1, overlap is recorded as 0.5, and no overlap is recorded as 0.

[0038] Extract the edge weights of the connections between candidate training action nodes and evaluation nodes in the rehabilitation knowledge graph. The edge weights represent the training effectiveness of the training action for this type of disorder in clinical experience. They are preset and normalized to the [0, 1] interval when the graph is constructed.

[0039] Based on the patient's functional level label in the assessment results, the range of motion intensity allowed for the current level is extracted and compared with the training intensity label of the candidate motion. If the intensity matches, the adaptation factor is assigned to 1; otherwise, the adaptation factor value is reduced according to the decreasing function.

[0040] The edge weight, muscle group matching degree index and function level adaptation factor are normalized item by item, and combined in a product form to obtain an adaptation degree weight of the candidate action, which is used to represent the comprehensive matching priority of the training action under the current patient state.

[0041] In a preferred embodiment, a personalized training path containing strength, frequency and rhythm parameters is constructed, and the specific process is as follows:

[0042] From the adaptation degree weight ranking list, a training action node set with an adaptation degree weight higher than a preset threshold is selected as a candidate action set of the training path;

[0043] For each training action node, a parameter template associated in the knowledge graph is extracted, and the parameter template includes recommended strength parameters, suggested frequency parameters and standard rhythm parameters of the training action;

[0044] The function level label in the evaluation result is matched with the training level requirement of the candidate training action, and if the current patient level is lower than the required level of the training action, the training action is excluded from the candidate set;

[0045] According to the patient's historical compliance record and the stage recovery curve, the recommended strength parameters, suggested frequency parameters and standard rhythm parameters of the candidate training action are adjusted to form target execution parameters under the current individual state;

[0046] According to the adaptation degree weight order of the training action node, the training actions with set execution parameters are arranged in sequence to form a personalized training path containing at least two training action nodes, and each action node in the personalized training path contains explicit strength parameters, frequency parameters and rhythm parameters.

[0047] In a preferred embodiment, interactive monitoring is performed simultaneously during the training execution process, and according to real-time action capture and electromyographic response recognition algorithm, training deviations are identified and prompt signals are output, and behavior completion degree and deviation level indicators are recorded, and the specific process is as follows:

[0048] At the beginning of the training action execution, the visual acquisition channel and the electromyographic signal acquisition channel are activated to obtain the action image sequence of the current training action and the corresponding electromyographic activation time sequence signal;

[0049] Based on the action image sequence, the displacement trajectories of the larynx and the head key points are extracted, and time sequence comparison is performed with the standard action template of the training action to identify spatial deviation behaviors such as posture deviation, start delay and insufficient amplitude;

[0050] The starting time, peak amplitude and duration of muscle activation are analyzed based on the electromyographic signal, and compared with the preset muscle group activation target parameters of the training action to identify electromyographic response deficiency, activation delay or abnormal tension and other electrophysiological deviation behaviors;

[0051] If any of the identified spatial deviation behaviors or electrophysiological deviation behaviors exceeds the set threshold, a prompt signal is triggered to output, and the prompt signal includes voice prompts, image prompts and vibration prompts;

[0052] After each training action is completed, the difference between the actual execution parameters and the target parameters of the training action is recorded, and the behavior completion degree of the action is calculated;

[0053] According to the cumulative number and deviation intensity level of the spatial deviation behavior and the electrophysiological deviation behavior, a deviation level index of the training action is generated and stored in the training behavior log as a progress evaluation reference.

[0054] In a preferred embodiment, the rehabilitation progress map is updated according to the training feedback record and the periodic score, and if the deviation level is continuously identified to exceed the limit, the training path is automatically reconstructed and the doctor terminal is synchronized to assist in strategy optimization and intervention. The specific process is as follows:

[0055] After each training action is completed, the corresponding behavior completion degree and deviation level index are recorded in the training feedback record, and the feedback data sequence of the current training period is constructed in chronological order;

[0056] At the end of the training period, the periodic score is calculated based on the feedback data sequence, and the periodic score is generated according to the total number of training actions, the average value of behavior completion degree and the distribution of deviation level, which is used to reflect the rehabilitation execution quality of the current stage;

[0057] According to the periodic score and the historical score curve, the progress state of the corresponding time node in the rehabilitation progress map is updated, and the rehabilitation progress map is a recovery trajectory structure constructed in chronological order, each node contains a stage score, an action compliance label and a deviation risk factor;

[0058] The number of times the deviation level exceeds the limit in the current training period is counted, and if the deviation level index of a plurality of consecutive training actions exceeds the preset threshold, the training path reconstruction mechanism is triggered;

[0059] In the training path reconstruction mechanism, the adaptation degree weight ordering process is called, and the adaptation degree weights are sorted from high to low according to the sorting results, the candidate training action nodes are reselected according to the sorting results, and a new training action sorting list is constructed;

[0060] The reconstructed training path and the current evaluation result are packaged and sent to the doctor terminal, and the doctor terminal receives the same and performs manual intervention, including adding or deleting training actions, adjusting parameter configuration or modifying rhythm factors, and finally synchronously updating to the patient execution end.

[0061] An auxiliary scheme generation system for swallowing disorder rehabilitation is used to implement the auxiliary scheme generation method for swallowing disorder rehabilitation.

[0062] A swallowing data acquisition module is used to acquire electromyographic signals, tongue pressure data, laryngeal movement images and speech features during swallowing based on a multi-channel sensing device, construct a multi-modal swallowing behavior feature vector, and uniformly map it into a standard time frame sequence.

[0063] A disorder stage recognition module is used to combine a pre-trained swallowing function evaluation model and historical rehabilitation data to perform disorder stage recognition and function level discrimination, and output a structured evaluation result containing disorder type, site and severity label.

[0064] A rehabilitation training path module is used to match relevant training action nodes in the rehabilitation knowledge graph according to the evaluation result, calculate the adaptation weight of the training action and generate an action sorting list, and construct a personalized training path containing intensity, frequency and rhythm parameters.

[0065] A training monitoring module is used to perform interactive monitoring during the training execution process, identify training deviations and output prompt signals according to real-time action capture and electromyographic response recognition algorithms, and record behavior completion degree and deviation level indicators.

[0066] A rehabilitation training optimization module is used to update the rehabilitation progress graph according to the training feedback record and the stage score, and if the deviation level is continuously identified to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronously updated to assist in strategy optimization and intervention.

[0067] The technical effects and advantages of the present application are as follows:

[0068] This invention synchronously collects electromyographic signals, tongue pressure data, laryngeal motion images, and acoustic response information during swallowing using multi-channel sensing devices. It then constructs a high-resolution, multimodal swallowing behavior feature vector through unified time alignment and feature normalization, accurately capturing the physiological response characteristics of each key stage in the entire swallowing process. Combined with a pre-trained swallowing function assessment model and a historical rehabilitation database, it performs stage identification and functional level judgment of swallowing disorders, outputting structured labels for disorder type, restricted area, and severity. Based on this, it matches highly suitable rehabilitation training action nodes using a knowledge graph reasoning mechanism, integrating edge... Personalized training pathways are constructed using weights, muscle group matching, and functional level adaptation factors, with clearly defined core parameters such as intensity, frequency, and rhythm to enhance the relevance and adherence of training plans. During training, deviation behaviors are identified through motion capture and electromyography (EMG) response linkage, triggering real-time prompts and recording the completion rate and deviation level of behaviors, achieving interactive monitoring and intelligent feedback throughout the process. Finally, the rehabilitation progress map is dynamically updated based on training feedback. If the deviation exceeds the limit, pathway reconstruction is triggered, and the physician's end is linked for strategy optimization and intervention, thereby improving the scientific nature, precision, and intelligence of rehabilitation, and significantly improving rehabilitation efficiency and safety. Attached Figure Description

[0069] Figure 1 This is a flowchart of a method for generating an auxiliary rehabilitation program for dysphagia according to the present invention.

[0070] Figure 2 This is a schematic diagram of the structure of an auxiliary scheme generation system for the rehabilitation of swallowing disorders according to the present invention. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] Example 1: As Figure 1 As shown, a method for generating an auxiliary rehabilitation program for dysphagia includes the following steps:

[0073] Based on the acquisition of electromyographic signals, tongue pressure data, laryngeal motion images and speech features during the swallowing process using multi-channel sensing devices, a multimodal swallowing behavior feature vector is constructed and uniformly mapped to a standard time frame sequence;

[0074] By combining a pre-trained swallowing function assessment model with historical rehabilitation data, the system performs obstacle stage identification and functional level discrimination, and outputs structured assessment results that include labels for obstacle type, location, and severity.

[0075] According to the evaluation results, the relevant training action nodes in the rehabilitation knowledge graph are matched, the adaptation weight of the training action is calculated, and an action ranking list is generated, and a personalized training path containing intensity, frequency and rhythm parameters is constructed;

[0076] During the training execution process, interactive monitoring is carried out synchronously, and according to real-time action capture and electromyographic response recognition algorithm, training deviation is recognized and prompt signal is output, and behavior completion degree and deviation level indicators are recorded;

[0077] According to the training feedback record and the stage score, the rehabilitation progress graph is updated, and if the deviation level is continuously identified to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronized to assist in strategy optimization and intervention.

[0078] Step 1, based on multi-channel sensing equipment, collect electromyographic signals, tongue pressure data, laryngeal movement images and speech characteristics during swallowing, construct multi-modal swallowing behavior feature vector, and uniformly map to standard time frame sequence, the specific implementation is:

[0079] The surface electromyography electrode array is arranged at the patient's lower jaw margin, above the larynx and sublingual muscle group position, the electrode is non-invasively connected in a skin-friendly manner, a high-sensitivity dual-channel electromyography preamplifier module is used to collect muscle contraction electrical signals during swallowing, the collection frequency is set to 1 kHz or higher, to ensure that the rapid electrical activity changes related to swallowing can be captured, the collected data is recorded in the form of time series to reflect the starting drive of the tongue and the protective action of the larynx, and is used as electromyographic signal data;

[0080] A flexible tongue pressure sensing pad is arranged on the upper palate and contacts the patient's tongue dorsum, and senses the contact pressure change between the tongue surface and the top of the oral cavity during swallowing. Through a high-precision strain pressure sensing unit, the pressure change value within a unit time is collected in real time, and the pressure duration of each time is recorded synchronously, so as to extract the intensity and rhythm characteristics of the tongue propulsion process, and serve as tongue pressure data;

[0081] An infrared camera arranged in front of the patient's neck obtains the vertical lifting track of the laryngeal prominence during swallowing, and uses a key point recognition algorithm between consecutive frames to extract the laryngeal prominence displacement, further calculates the lifting speed and maximum displacement amplitude, and captures the laryngeal closure and opening process through the data, reflects the action coordination of the food bolus passing through the esophageal inlet, and serves as laryngeal movement image data;

[0082] The close-range pickup microphone is placed on the neck side of the patient, and the laryngeal sound accompanying the swallowing action is captured in real time through the voice signal acquisition module. The acquisition content includes the sound starting point, peak frequency component and the entire acoustic event duration at the swallowing starting moment, and is subjected to filtering and noise reduction processing to exclude environmental interference, for analyzing the integrity and contraction rhythm of the glottal vibration, and serving as the voice feature data;

[0083] After the four types of heterogeneous sensing data, i.e., the myoelectric signal data, the tongue pressure data, the laryngeal movement image data and the voice feature data, are collected, a standardized time axis is constructed with the swallowing starting moment as the unified reference datum point, and the feature data of different modalities are mapped to the unified time frame structure. The structure aligns the original signals according to the relative time, normalizes the length, and generates a fixed-length, multi-modal joint vector for the input of the subsequent evaluation model, and finally forms a swallowing behavior feature vector set containing the myoelectric response curve, the tongue pressure peak value sequence, the laryngeal movement trajectory parameter and the acoustic signal time-frequency feature, which constitutes a complete swallowing behavior feature expression.

[0084] The pre-trained swallowing function evaluation model and the historical rehabilitation data are combined to perform the disorder stage recognition and the function grade discrimination, and output the structured evaluation results containing the disorder type, the position and the severity label. The specific implementation is as follows:

[0085] The swallowing behavior feature vector under the standard time frame structure is input into the swallowing function evaluation model. The swallowing evaluation model adopts an end-to-end neural network structure. The front end of the swallowing function evaluation model adopts a multi-channel convolution structure to extract the spatial distribution features of the electromyographic signal, tongue pressure change, image trajectory and acoustic parameters. In the middle layer, the features of each mode are fused by feature splicing. The back end of the model adopts a recurrent neural network structure (such as a bidirectional gated recurrent unit Bi-GRU or LSTM) to model the change trend of swallowing behavior in the time dimension and capture the dynamic evolution relationship between different action stages. An attention mechanism is introduced after the recurrent structure. The attention mechanism is based on the importance weight distribution strategy of the time sequence to weight and enhance the key frame section in the input sequence. The key frame section is determined by labeled samples or template matching. Specifically, it includes: swallowing initiation period (corresponding to the first rise of electromyography and the appearance of tongue pressure), laryngeal elevation period (corresponding to the vertical displacement peak of laryngeal trajectory), contraction recovery period (corresponding to the attenuation of electromyography and the fall of laryngeal prominence). The attention mechanism assigns higher weights to the feature frames in these periods, so that the model focuses on learning the feature differences in this stage during the training process, improving the accuracy of obstacle stage recognition. The output end of the model adopts a multi-classification structure to identify the obstacle stages existing in the swallowing process and output the corresponding obstacle type label. That is, after the swallowing function evaluation model processes the input swallowing behavior feature vector, a set of structured prediction results is output. The structured prediction results include the classification probability prediction of the obstacle stage, and the obstacle stage involved in the swallowing process is identified in the output structured prediction results. The obstacle stage division is based on the physiological process of swallowing action, mainly including oral preparation period obstacle (such as tongue propulsion weakness), pharyngeal period obstacle (such as laryngeal lifting delay) and esophageal transportation period obstacle (such as swallowing residue or transportation delay). Through a soft classification strategy, the probability of three types of obstacle labels is judged, and multiple label results can be output at the same time to reflect mixed obstacles.

[0086] The fine division of the function level is to match the feature vector output by the current model with the standard sample set in the historical rehabilitation database. The standard sample set is constructed by rehabilitation centers or clinical institutions. Each sample has complete function level labels (such as mild, moderate and severe), and contains multi-modal feature expressions consistent with the input dimensions of the current model. The similarity calculation is based on weighted Euclidean distance or Mahalanobis distance measurement methods, respectively comparing electromyographic response difference, tongue pressure peak, laryngeal movement trajectory and acoustic features, generating a ranking list, and selecting a reference sample set with a similarity higher than a set threshold from the ranking list. According to the frequency distribution of the function level labels in the reference sample, the function level range of the current evaluation object is determined.

[0087] The above recognition result is structured and output to generate an evaluation report. The structured evaluation result includes the following three types of fields: 1) an obstacle type field, reflecting the identified obstacle stage type; 2) a main restricted part field, extracting the muscle group part label involved in the restriction, such as the front part of the tongue, the pharyngeal constrictor muscle, or the glottis closing muscle group; and 3) a severity level field, marking the functional level grading result of the current patient, which can be annotated according to a five-level or three-level system for subsequent training path generation.

[0088] Specifically, the key frame section refers to the time segment corresponding to the three core action stages in the swallowing physiological process in the standardized time sequence covered by the entire swallowing behavior feature vector, including:

[0089] The swallowing initiation period key frame section starts from the oral cavity closing action to the time interval when the tongue pressure peak initially appears. In this section, the electromyography signal appears the first wave rise, the tongue pressure sensor detects significant pressure changes, and is accompanied by initial glottis vibration, which is the key trigger stage of the swallowing action;

[0090] The laryngeal elevation period key frame section corresponds to the stage of rapid laryngeal elevation. In this time period, the laryngeal movement trajectory presents the maximum vertical displacement, the laryngeal key point acceleration in the visual image is significantly enhanced, and the electromyography signal reaches the second wave peak, which is the main motion output stage in the swallowing process;

[0091] The contraction recovery period key frame section starts from the laryngeal movement stopping and downward recovery to the interval when the electromyography signal decays smoothly. In this section, it is usually accompanied by muscle relaxation and rapid weakening of electrophysiological activity, which is the transition stage between the completion of the swallowing action and the preparation for the next swallowing;

[0092] The start and end time points of each key frame section can be determined by using the following methods: using the change rate of electromyography signal, the derivative of tongue pressure signal, the speed threshold of laryngeal key point displacement, and the energy mutation point in the acoustic signal as the segmentation basis, presetting the section boundary time proportion in the standard swallowing template, dynamically mapping the current collected data event alignment benchmark, using dynamic time warping time alignment technology to find the optimal correspondence of the current sample and the standard template at the key feature points, and then marking the key frame section.

[0093] The swallowing function evaluation model prediction result is matched with the samples with clear function level labels in the historical rehabilitation database, and the similarity measurement method is used to confirm the range of the current function level. More specific steps include:

[0094] A standard sample set is extracted from a historical rehabilitation database, which is composed of a large number of historical patient data with completed rehabilitation process and clear functional level, each standard sample contains a complete multi-modal swallowing behavior feature vector, which is consistent with the feature extraction structure used by the current evaluation object, ensuring that the data dimensions and semantics are completely aligned, and each sample has a verified functional level label, which can adopt a five-level system (such as level one to level five) or a three-level system (mild, moderate, severe), according to the actual clinical standard setting; the multi-modal feature vector output by the current swallowing function evaluation model is preprocessed to ensure that its numerical range and unit scale are consistent with the standard sample set, and the normalization process is set according to the physical quantity unit of each modality: the maximum peak amplitude is taken as the normalization reference for electromyography signal, the upper limit of pressure is taken as the reference interval for tongue pressure data, the maximum displacement amplitude is standardized for laryngeal movement trajectory, and the time length and frequency value are normalized to the standard interval for acoustic features; the normalization of each modality is realized through linear proportional mapping, so that all feature components fall into a unified numerical interval, which can usually be set to [0, 1] or [-1, 1];

[0095] After normalization, each sample data in the standard sample set is traversed in turn, and the feature distance between it and the current evaluation feature is calculated, which is a comprehensive multi-modal difference measurement index, and the combination calculation is described as follows:

[0096] First, calculate the difference items of electromyographic response, including activation start time difference, peak amplitude difference and duration difference; second, extract the maximum pressure difference and pressure holding time difference in the tongue pressure data, then compare the maximum lifting displacement and descending time in the laryngeal key point movement trajectory to get the laryngeal trajectory difference, finally analyze the starting time offset, frequency bandwidth difference and total duration difference of acoustic features, weight and accumulate the feature differences in each modality to get the total difference value of this modality, that is, add all the modal difference values to get the final multi-modal feature distance;

[0097] After the feature distance calculation is completed, a similarity ranking list is generated based on the distance results of all samples. In order to filter out outliers, only samples with similarity higher than the set threshold are retained to form a reference sample set, the threshold can be a fixed distance value, or a percentage of the top samples (such as the top 10%) can be used for screening, in the example, at least two reference samples are retained to ensure the diversity and stability of functional level judgment;

[0098] Further statistics the functional grade labels corresponding to each sample in the reference sample set, and performs statistical analysis on the frequency of the labels. The functional grade with the highest frequency is taken as the grade judgment result of the current evaluation object, that is, the functional level of the current patient at the evaluation moment is considered to be closest to the state of the sample corresponding to the label. If there are multiple grade labels with the same frequency, a weighted voting method is introduced, and the label voting is weighted according to the similarity value of the sample. The higher the weight of the sample, the higher the priority of the grade label corresponding to the sample. The output result of the matching process is the functional grade label corresponding to the current patient, which is marked in the severity grade field as part of the structured evaluation result, for use in personalized training path selection and parameter adjustment.

[0099] According to the evaluation result, the related training action nodes in the rehabilitation knowledge graph are matched, the adaptation weight of the training action is calculated, and an action ranking list is generated. A personalized training path containing intensity, frequency and rhythm parameters is constructed, which is specifically implemented as:

[0100] The structured evaluation result output by the swallowing function evaluation module is received, and three core elements contained therein are extracted: disorder type label, main restricted muscle group position label and severity grade label. The above three labels constitute a set of to-be-matched nodes for action matching and are input into the rehabilitation knowledge graph as input;

[0101] The rehabilitation knowledge graph is modeled using a heterogeneous graph structure. The nodes in the rehabilitation knowledge graph include but are not limited to disorder type nodes, training action nodes, muscle group nodes, functional grade nodes and different semantic levels. The edges in the graph represent the association between nodes, especially the training adaptation edges between the disorder type nodes and the training action nodes. Each edge is provided with an edge weight parameter. The edge weight represents the adaptation degree of the training action for a certain disorder type in clinical experience or literature data. It is obtained by expert annotation, system learning or experience statistics, and the value range is usually normalized to [0, 1]. Starting from the disorder type node in the knowledge graph, a breadth-first search traversal operation is performed, and the maximum search layer is 2 hops to ensure that direct associated actions and training action nodes indirectly associated through muscle group nodes are captured. All training action nodes that meet the connectivity condition in the search process are collected as a training action candidate set. For each candidate training action node, the adaptation weight between it and the current evaluation result is calculated to determine its adaptation priority in the current training;

[0102] The calculation of the adaptation weight is based on three core factors:

[0103] The edge weight factor is to extract the association edge weight between the current training action node and the disorder type node, which reflects the historical effect or adaptation of the action in training the type of disorder;

[0104] The muscle group matching factor is matched with the restricted muscle group label extracted from the evaluation result according to the target muscle group acted on by the candidate training action, and if the training action completely covers the target muscle group, the value is 1; if it partially covers, the value is 0.5; if there is no intersection, the value is 0;

[0105] The function level adaptation factor is determined according to the function level label evaluated by the current patient, whether the training action meets the training intensity range allowed by the current level, if it completely matches, the value is 1; if it exceeds the current level requirement, a decreasing factor less than 1 is set according to the training level deviation, for example, 0.8, 0.5 or 0;

[0106] After normalizing the above three factors respectively, they are combined in a product manner to obtain the adaptation degree weight corresponding to each candidate training action node, that is, the adaptation degree weight is equal to the edge weight multiplied by the muscle group matching degree multiplied by the function level adaptation factor. This combination method can effectively suppress the inadaptation action when any one factor is zero, and improve the consistency of multi-dimensional adaptability expression ability.

[0107] All candidate training action nodes are sorted according to the adaptation degree weight from high to low to generate a training action sorting list, and the actions in the sorting list will be used as the candidate action set when building the personalized training path, and the training action nodes with higher adaptation degree weight are preferred to ensure that each training task in the training path is highly matched with the current state of the patient, and the efficiency and pertinence of rehabilitation training are improved.

[0108] The specific process of building a personalized training path containing strength, frequency and rhythm parameters is as follows:

[0109] Select the training action nodes with weight value higher than the set threshold value from the adaptation degree weight sorting list to form a training action candidate set, and set the threshold value as the minimum adaptation requirement value that ensures training safety and effectiveness under the current function state;

[0110] For each training action node in the training action candidate set, extract the parameter template associated with it in the preset rehabilitation knowledge graph, which includes: training intensity parameter (indicating muscle group load level during action execution), training frequency parameter (indicating recommended number of repetitions per unit period) and training rhythm parameter (indicating time rhythm and control beat of each execution);

[0111] Extract the function level label output in the evaluation result, and compare the label with the training level requirement of the candidate training action node, if the current patient function level is lower than the minimum level requirement of the training action, the training action node is removed, and the candidate set is updated;

[0112] Obtaining the historical compliance record of the patient, the compliance record including the completion frequency of past training tasks, the action deviation level and the training interruption condition; at the same time, the stage recovery curve of the patient is extracted, including the functional score trend and the deviation correction feedback result;

[0113] According to the training tolerance and adaptation trend reflected in the compliance record and the recovery curve, the parameter template of each candidate training action is adaptively adjusted, and the adjustment mode includes: when the compliance is poor or the recovery is slow, the training intensity parameter and the rhythm parameter are reduced; when the recovery stability is good and the functional level is gradually improved, the frequency and rhythm parameters are appropriately increased, the target execution parameter group under the individual state is formed, the training action nodes with completed parameter setting are arranged in order from high to low according to the adaptation degree weight of the training action nodes, and the personalized training path is constructed and formed, the personalized training path is composed of at least two training action nodes, and each training action node includes determined training intensity parameter, training frequency parameter and training rhythm parameter, which is used to guide the specific execution of the subsequent training plan.

[0114] During the training execution process, interactive monitoring is carried out synchronously, the training deviation is identified according to the real-time action capture and electromyographic response identification algorithm, and the prompt signal is output, and the behavior completion degree and the deviation level index are recorded, and the specific implementation is:

[0115] Before the start of the training action in each round of the personalized training path, the visual acquisition device and the electromyographic signal acquisition device are started respectively, the visual acquisition device is used to continuously capture the image sequence of the neck region, the throat profile and the head posture of the patient, the electromyographic acquisition device is used to record the electromyographic activation response of the target muscle group during the training process, both are synchronized and aligned with the training starting time as the unified reference, the image data is extracted through the key point identification algorithm to obtain the spatial displacement trajectory of the throat and the head, and is compared with the preset standard training action template frame by frame to identify the spatial deviation of the displacement trajectory in terms of angle change, motion amplitude and speed; at the same time, the electromyographic response signal is analyzed into three dimensions: the starting time point of muscle activation, the peak amplitude value and the activation duration, and is compared with the preset muscle group activation target range of the training action item by item to identify whether there is early activation, insufficient peak value or insufficient activation and other electrophysiological deviations;

[0116] The process of setting the deviation recognition threshold standard is: if any one of the spatial deviation or the electrophysiological deviation exceeds the set error limit, a visual or audible prompt signal is triggered immediately to prompt the patient to adjust the posture or force output; the prompt signal type is bound to the corresponding deviation type, so that the feedback has a clear guiding significance. After each training action is completed, the actual execution parameters are recorded, including the actual action duration, the actual peak displacement, and the electromyographic activation characteristics, etc. The difference between the actual parameters and the target parameters set in the current training action node is calculated to obtain the behavior completion degree index of the action. The closer the behavior completion degree value is to 1, the more accurate the execution is. The system also counts the number of spatial deviation behaviors, the number of electrophysiological deviation behaviors, and the amplitude level during each training process, and uses a weighted integral method to calculate the comprehensive deviation level index. For example, if the spatial deviation occurs twice, the electrophysiological deviation occurs once, and the intensity level is moderate, the corresponding deviation level index is generated as a moderate level. The completion degree and deviation level data in all training processes are stored in the training behavior log in real time, which is used as the basis for rehabilitation progress evaluation and subsequent path reconstruction.

[0117] For example, during a throat lifting training process, the action capture and electromyographic signal acquisition channels are simultaneously turned on. The camera records that the patient's laryngeal prominence lifting amplitude is only 60% of the standard template, the rising delay is 0.8 seconds, and the dwell time is less than 0.5 seconds, which is significantly lower than the training requirements. At the same time, the electromyographic sensor collects the peak value of the target muscle group activation signal, which is only 50% of the preset target, and the duration is short, with early attenuation phenomenon. The system identifies the dual deviation behaviors of insufficient lifting amplitude and insufficient muscle activation intensity according to the comparison results of spatial displacement and electromyographic response. According to the comparison results of spatial displacement and electromyographic response, the system triggers voice and interface prompts to inform the patient that the throat lifting amplitude and maintenance time are low, and prompts the patient to pay attention to the action quality. After the training is completed, the system compares the actual and target parameters to generate a behavior completion degree of 0.55, and evaluates the deviation level of the current action as 2 based on the deviation item accumulation, which is written into the periodic feedback record as part of the training log.

[0118] According to the training feedback record and the stage score, the rehabilitation progress map is updated. If the deviation level is continuously identified to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronized to assist in strategy optimization and intervention. The specific implementation is:

[0119] After each training action is performed, the behavior completion degree index and the deviation level index of the current action are automatically recorded and sequentially stored in time order to form a training feedback data sequence, which covers the execution results of all training actions in the current training period. The behavior completion degree index represents the execution quality of each training action, which is usually normalized by calculating the difference between the target parameter and the actual parameter. The deviation level index is a numerical label obtained by grading based on the frequency and amplitude of spatial deviation and electrophysiological deviation. After the training period ends, the system performs statistical analysis on the feedback data sequence in the period to calculate the stage score of the current stage. The stage score is generated according to the following rules: taking the total number of training actions in the current period as the base, combining the average value of the behavior completion degree of all training actions, and weighting the deviation level distribution, for example, multiplying the number of moderate deviations by a weight of 0.5 and multiplying the number of severe deviations by a weight of 1, to generate a stage execution score (the numerical range can be normalized to [0, 100]). The score reflects the overall quality level of the rehabilitation task execution.

[0120] The stage score result is mapped to a rehabilitation progress graph. The graph is a trajectory structure constructed based on time series. Each time node represents a comprehensive state node of a training period, including the stage score value, the training adherence label (such as high, medium, and low), and the deviation risk factor (calculated by accumulating the frequency and severity of the deviation level). The number of times the deviation level index exceeds the set threshold in the current training period is counted. If the deviation level of multiple consecutive training actions exceeds the preset threshold (such as 3 consecutive times of high deviation level), it is determined that the current training path has a mismatch risk, and the training path reconstruction mechanism is automatically triggered.

[0121] After triggering the reconstruction mechanism, the system re-invokes the aforementioned evaluation result and knowledge graph matching process, recalculates the adaptation weight of the candidate training action node based on the current functional level label of the patient and the training adherence record, and preferentially selects action nodes with high adaptability, moderate training intensity, and good historical adherence to generate a new training action sorting list. The list forms a new training path, and the path execution parameter package (including the intensity, frequency, and rhythm parameters of each action) is generated based on the current evaluation result. The path package is automatically sent to the doctor terminal, and the doctor can perform intervention operations on the training path in the terminal, such as replacing training actions, adjusting rhythm control, resetting training intensity or rhythm upper and lower limits, etc. After the doctor confirms the reconstructed path, it is updated to the patient execution terminal in real time to realize real-time optimization of the path and closed-loop adjustment of the rehabilitation strategy. This process ensures that the training feedback and doctor intervention form a linkage mechanism, improving the adaptability and safety of the training.

[0122] It should be noted that the threshold values involved in the embodiments can be determined according to specific scenarios and requirements.

[0123] The application synchronously collects myoelectric signals, tongue pressure data, laryngeal movement images and acoustic response information in the swallowing process through a multi-channel sensing device, constructs a high-resolution multi-modal swallowing behavior feature vector through unified time alignment and feature normalization processing, accurately captures the physiological response characteristics of each key stage in the whole swallowing process, combines a pre-trained swallowing function evaluation model and a historical rehabilitation database, performs stage recognition and function grade discrimination of swallowing disorders, and outputs structured disorder type, restricted site and severity label; on this basis, based on the knowledge graph reasoning mechanism, the rehabilitation training action node with high adaptation degree is matched, the edge weight, muscle group matching degree and function grade adaptation factor are fused to construct a personalized training path, and the core parameters such as intensity, frequency and rhythm are clearly configured, thereby enhancing the pertinence and compliance of the training plan; in the training execution process, the deviation behavior is recognized through motion capture and myoelectric response linkage, real-time prompts are triggered, and behavior completion and deviation grade are recorded, realizing interactive monitoring and intelligent feedback in the whole process; finally, the rehabilitation progress graph is dynamically updated according to the training feedback, and if the deviation is out of limit, the path reconstruction is triggered and the doctor terminal is linked to optimize and intervene the strategy, thereby improving the scientific nature, accuracy and intelligent level of rehabilitation, and significantly improving the rehabilitation efficiency and safety.

[0124] Embodiment 2: An auxiliary scheme generation system for swallowing disorder rehabilitation, as shown in Figure 2 specifically comprises:

[0125] A swallowing data acquisition module is configured to acquire myoelectric signals, tongue pressure data, laryngeal movement images and speech features in the swallowing process based on a multi-channel sensing device, construct a multi-modal swallowing behavior feature vector, and uniformly map it to a standard time frame sequence;

[0126] A disorder stage recognition module is configured to combine a pre-trained swallowing function evaluation model and historical rehabilitation data, perform disorder stage recognition and function grade discrimination, and output structured evaluation results including disorder type, site and severity label;

[0127] A rehabilitation training path module is configured to match related training action nodes in the rehabilitation knowledge graph according to the evaluation results, calculate the adaptation degree weight of the training action and generate an action sorting list, and construct a personalized training path containing intensity, frequency and rhythm parameters;

[0128] A training monitoring module is configured to perform interactive monitoring during the training execution process, identify training deviations and output prompt signals according to real-time motion capture and myoelectric response recognition algorithms, and record behavior completion and deviation grade indicators;

[0129] A rehabilitation training optimization module is configured to update the rehabilitation progress graph according to the training feedback record and the stage score, and if the deviation grade is continuously identified to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronized to assist strategy optimization and intervention.

[0130] The above-described embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, an ATA hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0131] It should be understood that the size of the sequence number of each process described above in various embodiments of the present application does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0132] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be realized in electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0133] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the embodiments of the device described above are merely schematic, and the division of the units is merely logical function division. There can be other division manners in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0134] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, and can be located in one position, or can be distributed on a plurality of network units.

[0135] In addition, the various functional units in the embodiments of the present application can be integrated in a processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in a unit.

[0136] The above describes only the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for generating an auxiliary program for the rehabilitation of dysphagia, characterized in that, The method comprises the following steps: Based on the multi-channel sensing device, the swallowing process is collected, including the myoelectric signal, the tongue pressure data, the laryngeal movement image and the voice feature, the multi-modal swallowing behavior feature vector is constructed, and is uniformly mapped into the standard time frame sequence; Combined with the pre-trained swallowing function evaluation model and the historical rehabilitation data, the obstacle stage recognition and function grade discrimination are performed, and the structured evaluation results containing the obstacle type, the position and the severity label are outputted; According to the evaluation results, the related training action nodes in the rehabilitation knowledge graph are matched, the adaptation degree weight of the training action is calculated, and the action sorting list is generated, and the individualized training path containing the intensity, the frequency and the rhythm parameters is constructed; During the training execution process, interactive monitoring is carried out synchronously, the training deviation is recognized according to the real-time action capture and the myoelectric response recognition algorithm, and the prompt signal is outputted, and the behavior completion degree and the deviation grade index are recorded; According to the training feedback record and the stage score, the rehabilitation progress graph is updated, if the deviation grade is continuously recognized to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronously assisted to optimize and intervene; Combined with the pre-trained swallowing function evaluation model and the historical rehabilitation data, the obstacle stage recognition and function grade discrimination are performed, and the structured evaluation results containing the obstacle type, the position and the severity label are outputted, and the specific process is as follows: The swallowing behavior feature vector constructed is inputted into the pre-trained swallowing function evaluation model, the swallowing function evaluation model comprises a convolution structure for extracting static features and a recurrent neural network structure for modeling time sequence changes; The response ability of the key frame section of the swallowing function evaluation model is enhanced using an attention mechanism, and the feature differences in the swallowing initiation period, the laryngeal elevation period and the contraction recovery period are determined; After the swallowing function evaluation model processes the input swallowing behavior feature vector, a set of structured prediction results is outputted, the structured prediction results comprise the classification probability prediction of the obstacle stage, the obstacle stage involved in the swallowing process is identified in the output structured prediction results, including the oral preparation period, the pharyngeal period and the esophageal transportation period, and the corresponding obstacle type label is outputted; The swallowing function evaluation model prediction result is matched with the samples with clear function grade labels in the historical rehabilitation database, and the current function grade range is confirmed through a similarity measurement method; Combined with the swallowing function evaluation model recognition result and the historical score mapping relationship, the structured evaluation result is generated, and the evaluation result comprises the obstacle type, the main limited position and the severity grade label; According to the evaluation results, the related training action nodes in the rehabilitation knowledge graph are matched, the adaptation degree weight of the training action is calculated, and the action sorting list is generated, and the individualized training path containing the intensity, the frequency and the rhythm parameters is constructed, and the specific process is as follows: Based on the structured evaluation result, the obstacle type, the limited position and the severity grade label are extracted and inputted into the pre-constructed rehabilitation knowledge graph as a set of nodes to be matched; The rehabilitation knowledge graph is a heterogeneous graph structure, and the graph comprises associated edges between the training action nodes and the obstacle type nodes, and the edge weight represents the adaptation degree of the action to a specific obstacle; In the rehabilitation knowledge graph, the obstacle type node is taken as the starting point, the breadth-first search is performed on the adjacent action nodes, and all directly connected or indirectly connected candidate training action nodes are extracted; For each candidate training action node, calculate the fitness weight between the candidate training action node and the evaluation result, which is determined by the edge weight, the muscle group matching degree of the action, and the fitness factor of the current functional level of the patient. Sort all candidate training action nodes according to the fitness weight from high to low to generate an action sorting list as the input candidate set for training path construction.

2. The method of claim 1, wherein the method further comprises: Based on the multi-channel sensing device, the electromyographic signals, tongue pressure data, laryngeal movement images and voice features during swallowing are collected, and a multi-modal swallowing behavior feature vector is constructed and uniformly mapped into a standard time frame sequence. The specific process is as follows: Through the surface electromyography electrode array, the time sequence electromyographic voltage signals generated by the muscle groups in the mandible, larynx and sublingual area during swallowing are obtained and recorded as electromyographic signal data. Through the tongue pressure sensing pad, the unit pressure change value of the tongue and the palate contact area is obtained, and the pressure duration is recorded synchronously, which is recorded as tongue pressure data. An infrared vision camera is used to obtain the vertical lifting trajectory of the laryngeal prominence during swallowing, and the lifting speed and amplitude are calculated, which are recorded as laryngeal movement image data. A close-range pickup microphone is used to collect the acoustic response signal during swallowing, and the starting time, peak frequency and duration are extracted, which are recorded as voice feature data. The electromyographic signal data, tongue pressure data, laryngeal movement image data and voice feature data are uniformly mapped into a standardized time frame structure with the swallowing event start as the reference point to form a swallowing behavior feature vector set.

3. The method of claim 2, wherein: The prediction results of the swallowing function evaluation model are matched with the samples with clear functional level labels in the historical rehabilitation database, and the current functional level range is confirmed by similarity measurement method. The specific process is as follows: Extract the standard sample set containing the functional level label from the historical rehabilitation database. Each sample in the standard sample set includes a swallowing feature vector of the same modal dimension as the current patient and its corresponding verified functional level label. Perform feature normalization processing on the feature vector output by the current swallowing function evaluation model. Calculate the feature distance between the normalized current feature vector and each sample in the standard sample set in turn. The feature distance calculation is based on the cumulative results of the numerical value differences between the modalities, including electromyographic response difference, tongue pressure peak difference, laryngeal trajectory change difference and acoustic duration difference. Generate a sample similarity sorting list from all distance calculation results, and select at least two samples with a similarity higher than a set threshold as reference samples. Statistically analyze the functional level label distribution in the reference samples, and determine the functional level range of the current evaluation object according to the highest frequency label as the final output of the severity determination result.

4. The method of claim 3, wherein: For each candidate training action node, calculate the fitness weight between the candidate training action node and the evaluation result, which is determined by the edge weight, the muscle group matching degree of the action, and the fitness factor of the current functional level of the patient. The specific process is as follows: Obtain the main limited muscle group information in the evaluation result, and compare it with the target muscle group involved in the candidate training action node to calculate a muscle group matching degree index. The muscle group matching degree index is quantified according to the target muscle group overlap degree, and is recorded as 1 for complete coverage, 0.5 for existing overlap, and 0 for no intersection; Extract the edge weight between the candidate training action node and the evaluation node in the rehabilitation knowledge graph, which represents the training effectiveness of the training action for the disorder in clinical experience and is normalized to the [0, 1] interval by default during graph construction; According to the functional level label of the patient in the evaluation result, extract the action intensity range allowed by the current level, and compare it with the training intensity label of the candidate action. If the intensity matches, assign an adaptation factor of 1, otherwise reduce the adaptation factor value according to a decreasing function; Normalize the edge weight, muscle group matching degree index, and functional level adaptation factor, and combine them in the form of product to obtain the adaptation degree weight of the candidate action, which represents the comprehensive matching priority of the training action under the current patient state.

5. The method of claim 4, wherein: Construct a personalized training path containing intensity, frequency, and rhythm parameters, with the following specific process: Select the training action node set with an adaptation degree weight higher than a preset threshold from the adaptation degree weight ordering list as the candidate action set of the training path; For each training action node, extract the associated parameter template in the knowledge graph, which includes the recommended intensity parameter, suggested frequency parameter, and standard rhythm parameter of the training action; Match the functional level label in the evaluation result with the training level requirement of the candidate training action. If the current patient level is lower than the required level of the training action, remove the training action from the candidate set; Adjust the recommended intensity parameter, suggested frequency parameter, and standard rhythm parameter of the candidate training action according to the patient's historical compliance record and the stage recovery curve to form the target execution parameter under the current individual state; According to the order of the adaptation degree weight of the training action node, arrange the training actions with set execution parameters in sequence to form a personalized training path containing at least two training action nodes, each action node in the personalized training path contains clear intensity parameter, frequency parameter, and rhythm parameter.

6. The method of claim 5, wherein: Synchronize interaction monitoring during training execution, identify training deviations and output prompt signals based on real-time motion capture and electromyographic response recognition algorithms, and record behavior completion degree and deviation level indicators, with the following specific process: At the beginning of training action execution, activate the visual acquisition channel and electromyographic signal acquisition channel to obtain the action image sequence and corresponding electromyographic activation time sequence signal of the current training action; Extract the displacement trajectory of the larynx and head key points based on the action image sequence, and compare it with the standard action template of the training action to identify spatial deviation behavior; Analyze the start time, peak amplitude, and duration of muscle activation based on electromyographic signals, and compare them with the preset muscle group activation target parameters of the training action to identify electrophysiological deviation behavior; If any of the identified spatial deviation behavior or electrophysiological deviation behavior exceeds the set threshold, trigger the prompt signal output; After each training action is completed, the difference between the actual execution parameter and the target parameter of the training action is recorded, and the behavior completion degree of the training action is calculated; According to the cumulative number and deviation intensity level of the spatial deviation behavior and the electrophysiological deviation behavior, a deviation level index of the training action is generated and stored in the training behavior log as a progress evaluation reference.

7. A method of generating a swallowing rehabilitation assistance protocol according to claim 6, wherein: According to the training feedback record and the periodic score, the rehabilitation progress graph is updated, and if the deviation level is continuously identified to be out of limit, the training path is automatically reconstructed and the doctor terminal is synchronized to assist strategy optimization and intervention. The specific process is as follows: After each training action is completed, the corresponding behavior completion degree and deviation level index are recorded in the training feedback record, and a feedback data sequence of the current training period is constructed in chronological order; At the end of the training period, the periodic score is calculated based on the feedback data sequence. The periodic score is generated according to the total number of training actions, the average value of behavior completion degree and the distribution of deviation level, which is used to reflect the rehabilitation execution quality of the current stage; According to the periodic score and the historical score curve, the progress state of the corresponding time node in the rehabilitation progress graph is updated. The rehabilitation progress graph is a recovery trajectory structure constructed in chronological order, and each node contains a stage score, an action compliance label and a deviation risk factor; The number of times that the deviation level exceeds the limit in the current training period is counted. If the deviation level index of the continuous training action exceeds the preset threshold, the training path reconstruction mechanism is triggered. In the training path reconstruction mechanism, the adaptation weight ordering process is called, and the adaptation weights are sorted from high to low. The candidate training action nodes are reselected according to the sorting, and a new training action sorting list is constructed. The reconstructed training path and the current evaluation result are packaged and sent to the doctor terminal. After receiving, the doctor terminal performs manual intervention, and the intervention result is updated to the patient execution end.

8. A system for generating a swallowing disorder rehabilitation assistance program for implementing the method of any one of claims 1-7, characterized in that, It includes: Swallowing data acquisition module, for collecting electromyographic signals, tongue pressure data, laryngeal movement images and voice features during swallowing based on multi-channel sensing devices, constructing multi-modal swallowing behavior feature vectors, and uniformly mapping them into standard time frame sequences; Disorder stage identification module, for combining pre-trained swallowing function evaluation model and historical rehabilitation data to perform disorder stage identification and function level discrimination, and outputting structured evaluation results containing disorder type, site and severity label; Rehabilitation training path module, for matching relevant training action nodes in the rehabilitation knowledge graph according to the evaluation results, calculating the adaptation weight of the training action and generating the action sorting list, and constructing the individualized training path containing intensity, frequency and rhythm parameters; Training monitoring module, for synchronously monitoring the training execution process, identifying training deviations and outputting prompt signals according to real-time action capture and electromyographic response recognition algorithm, and recording behavior completion degree and deviation level index; Rehabilitation training optimization module, for updating the rehabilitation progress graph according to the training feedback record and the periodic score, and automatically reconstructing the training path and synchronizing the doctor terminal to assist strategy optimization and intervention if the deviation level is continuously identified to be out of limit.

Images