Multi-mode swallowing function intelligent evaluation and intervention integrated system

Abstract

The invention discloses a multi-modal swallowing function intelligent evaluation and intervention integrated system, and relates to the technical field of intelligent medical rehabilitation, and the system comprises a multi-modal data collection module which collects images, physiological signals and clinical scale data and associates patient identification storage; the multi-modal intelligent evaluation module quantifies tongue bone displacement, UES open area and FA value through image analysis, calculates myoelectricity incubation period and throat uplift time through physiological analysis, and outputs obstacle grade and risk grade in combination with machine learning; the acupuncture therapy evaluation and adaptation module is used for selecting acupuncture points according to a swallowing period and configuring operation parameters; the combined rehabilitation intervention module comprises glacial acid stimulation, standard swallowing training and neuromuscular electrical stimulation; and the result feedback and system iteration module is used for generating a report, encrypting storage data and iterating the model. According to the invention, dysphagia can be accurately evaluated, acupuncture and rehabilitation schemes are adapted according to grading and periods, the incidence rate of aspiration pneumonia is effectively reduced, the swallowing function and nutrition state are improved, and the rehabilitation efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent medical rehabilitation, and particularly relates to a multi-modal swallowing function intelligent evaluation and intervention integrated system. BACKGROUND

[0002] Swallowing disorder is a common complication of nervous system diseases, head and neck tumors and senile degenerative diseases. Its clinical management has long been faced with the dual challenges of evaluation fragmentation and intervention homogeneity. Although the current clinical evaluation tools have their own advantages, they lack a multi-dimensional integration mechanism. The clinical scoring system centered on the DOSS-S scale has been proven to have excellent reliability (such as test-retest reliability and inter-rater reliability), and has a strong correlation with the gold standard VFSS and FOIS scale. However, its main function is limited to the classification of the severity of swallowing disorders, and it cannot simultaneously correlate with pathological features in the imaging field, such as the reduced opening cross-sectional area of the upper esophageal sphincter (UES) caused by the dorsal lateral medulla injury, the reduced anisotropy fraction (FA value) corresponding to the white matter microstructure damage of the corticobulbar tract, and other key indicators, resulting in insufficient correlation between the evaluation results and the potential neuroanatomical or physiological pathological mechanisms. At the same time, although the video fluoroscopy swallowing study (VFSS) can quantify dynamic parameters such as hyoid displacement and pharyngeal transit time, and the fiberoptic endoscopic swallowing function test (FEES) can observe pharyngeal mucosa and food residue, such image evaluation is usually carried out independently and does not form a closed loop with clinical data such as the DOSS-S scale and the NRS2002 nutritional risk score, making it difficult to fully reflect the synergistic relationship between the patient's swallowing function, nutritional status and aspiration risk.

[0003] The lack of individualization of intervention programs further restricts the clinical efficacy. Existing acupuncture treatment has been proven to be effective for dysphagia after stroke, such as the stage acupuncture method which selects acupoints according to the oral stage, swallowing stage and esophageal stage. However, the staging criteria are not deeply bound to objective evaluation indicators, and only rely on simple screening methods such as the Katakura drinking water test, which cannot adjust acupoints and operation parameters according to quantitative indicators such as DOSS-S classification and UES opening state, resulting in insufficient difference in intervention programs for patients with severe disorders (DOSS-S 1-2) and mild disorders (DOSS-S 5-7). In combined rehabilitation intervention, ice acid stimulation can improve swallowing function by enhancing the neural plasticity of the solitary nucleus-insular pathway, and needs to be standardized according to the concentration gradient (4%→8%) and low temperature (4℃). However, the existing scheme does not dynamically adjust the stimulation frequency and duration in combination with real-time evaluation results, and is not suitable for patients with dorsal lateral medulla injury and other special pathological types; standard swallowing training includes oral movement, pharyngeal stimulation and other content, but lacks linkage adjustment with image evaluation and physiological signals, making it difficult to solve the core problems such as the collapse of the circopharyngeal muscle coordination and the delay of the laryngeal elevation.

[0004] Furthermore, there are significant shortcomings in the full-cycle monitoring and data utilization of dysphagia management. In clinical practice, aspiration risk monitoring relies heavily on manual observation of symptoms such as coughing and fluctuations in blood oxygen saturation, failing to provide early warnings through real-time physiological signals and imaging features. This limits the effectiveness of pneumonia prevention in high-risk patients. Simultaneously, assessment data and intervention plans from different medical institutions are often stored independently, lacking standardized data synchronization and sharing mechanisms. This hinders the provision of sufficient multi-center data for iterative machine learning models, restricting the continuous optimization of assessment and intervention plans. For example, predictive models based on pharyngeal constrictor muscle pressure and insular ReHo values ​​are difficult to generalize and cannot provide more precise decision support for individualized rehabilitation. Ultimately, this results in inconsistent treatment outcomes for patients with dysphagia, with both the recovery rate of oral feeding function and the prevention of aspiration-related pneumonia failing to reach ideal levels. Summary of the Invention

[0005] The present invention proposes an integrated system for intelligent assessment and intervention of multimodal swallowing function to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a multimodal intelligent assessment and intervention system for swallowing function, comprising the following modules: The multimodal data acquisition module collects swallowing function images, physiological signals, and clinical scale data. Image data is acquired through video-assisted fluoroscopy swallowing examination, fiberoptic endoscopy swallowing function examination, and 3.0 T MRI equipment. Physiological signal data is acquired through surface electromyography and ultrasound equipment. Clinical scale data is entered through the system's interactive interface. All data is associated with a unique patient identifier and stored in DICOM, EDF, and JSON formats. The multimodal intelligent assessment module comprises four units: image analysis, physiological signal analysis, scale scoring, and machine learning fusion. The image analysis unit calculates the hyoid bone displacement amplitude, UES open peak area, and corticobulbar fasciculus FA value using algorithms. The physiological signal analysis unit calculates the cortical-laryngoscopy transmission latency and laryngeal elevation time. The scale scoring unit grades the DOSS-S scale. The machine learning fusion unit uses the XGBoost algorithm to output the degree of swallowing disorder, aspiration risk probability, and nutritional risk level. The model training uses 5-fold cross-validation. The acupuncture treatment assessment and adaptation module includes four units: stage assessment, acupoint selection, operation parameter configuration, and effect assessment. The stage assessment unit divides swallowing disorders into oral, swallowing, and esophageal stages. The acupoint selection unit selects corresponding acupoints according to the stage. The operation parameter configuration unit sets the needle insertion depth, needle retention time, and special operations for acupoints in each stage. The effect assessment unit evaluates the effect through the Kubota water swallowing test and standard scales. The combined rehabilitation intervention module includes three units: ic acid stimulation, standard swallowing training, and neuromuscular electrical stimulation. The ic acid stimulation unit is equipped with a citric acid concentration gradient and sets the stimulation temperature, location, duration, and frequency. The standard swallowing training unit includes oral and pharyngeal stimulation and feeding adjustment training. The neuromuscular electrical stimulation unit uses NMES technology and sets the electrode position, frequency, pulse width, current intensity, and training duration and frequency. The results feedback and system iteration module generates multimodal assessment and intervention reports, records assessment data after each intervention, and stores it in a distributed database using AES256 encryption. The machine learning fusion model is iterated by adding new data, retrained every quarter, and an emergency model iteration is triggered when the accuracy drops by more than 5%.

[0007] Furthermore, it also includes a comprehensive swallowing function scoring module, which uses a formula... Calculate the patient's overall swallowing function score, where S represents the overall swallowing function score. As image evaluation weight, Weighting of physiological signals For the scale scores, and The scores were determined using a machine learning model trained on data from 3,000 historical patients; S1 was the imaging assessment score, S2 was the physiological signal assessment score, and S3 was the scale score.

[0008] Furthermore, it includes a multimodal data quality verification module. This module verifies the quality of acquired image data and physiological signal data. Image data verification indicators include the clarity of video-assisted swallowing examination, the mucosal visibility of fiberoptic endoscopy swallowing function examination, and the signal-to-noise ratio of MRI DTI images. For video-assisted swallowing examination, the image clarity requires an intra-frame contrast agent boundary clarity ≥90%. If it is lower than this value, it prompts re-acquisition and adjustment of the device focus. For fiberoptic endoscopy swallowing function examination, the mucosal visibility requires a mucosal texture recognition ≥85%. If it is lower than this value, it prompts cleaning of the endoscope lens. The signal-to-noise ratio of MRI DTI images requires ≥20dB. If it is lower than this value, the scan time is extended. Physiological signal data verification indicators include the signal-to-noise ratio of surface electromyography and the resolution of ultrasound images. The signal-to-noise ratio of surface electromyography requires ≥30dB. If it is lower than this value, the electrode position and skin impedance are adjusted. The resolution of ultrasound images requires a hyoid-thyroid cartilage structure resolution ≥95%. If it is lower than this value, the probe position and frequency are adjusted. Data that passes the verification enters the evaluation module, while data that fails the verification is marked with the reason and guided to re-acquisition.

[0009] Furthermore, the acupuncture treatment assessment and adaptation module also includes an acupoint positioning and navigation unit. This unit uses ultrasound guidance to locate acupoints. During the oral cavity phase, the ultrasound probe is placed in the anterior midline of the neck to display the soft tissue layer below the hyoid bone corresponding to the Lianquan acupoint, marking the needle insertion point and depth. During the swallowing phase, when locating the Fengchi acupoint, the ultrasound probe is placed below the occipital bone at the back of the neck to display the soft tissue lateral to the foramen magnum corresponding to the Fengchi acupoint, determining the needle insertion direction and angle. During the esophageal phase, when locating the esophageal acupoint via auricular acupuncture, a high-frequency ultrasound probe is placed in the concha cavity area to display the junction of cartilage and soft tissue corresponding to the esophageal acupoint, marking the puncture point. Simultaneously, the ultrasound images during the needle insertion process are displayed in real time.

[0010] Furthermore, it also includes a rehabilitation outcome prediction module, which uses formulas... Predict the probability of effective rehabilitation intervention, where P is the probability of effective intervention and t is the duration of intervention. The difference between the comprehensive swallowing function score before and during the intervention is represented by a, where a is the time coefficient, b is the score change coefficient, and c is a constant. When P ≥ 0.8, the intervention plan is considered suitable, and the current plan is continued. When 0.5 ≤ P < 0.8, the intervention parameters are adjusted. When P < 0.5, the intervention plan is reassessed and changed.

[0011] Furthermore, the combined rehabilitation intervention module also includes a dynamic monitoring unit for aspiration risk. This unit collects physiological signals of the patient during swallowing in real time, including the swallowing reflex latency from surface electromyography, the laryngeal elevation delay time from ultrasound monitoring, and blood oxygen saturation. When the swallowing reflex latency exceeds 200ms, the laryngeal elevation delay time exceeds 400ms, or the blood oxygen saturation drops below 94%, a real-time warning is triggered, and medical staff are notified through system interface prompts and audible alarms.

[0012] Furthermore, it also includes a nutrition support adaptation module, which configures a nutrition plan based on the NRS2002 nutritional risk score and serum prealbumin level. When the NRS2002 score is ≥3 or the serum prealbumin is <180mg / L, it is judged as high nutritional risk, and the daily nutritional intake is calculated as body weight × 35kcal; when the NRS2002 score is 1-2 and the serum prealbumin is 180-220mg / L, it is judged as moderate nutritional risk, and the daily nutritional intake is calculated as body weight × 30kcal; when the NRS2002 score is 0 and the serum prealbumin is ≥220mg / L, it is judged as low nutritional risk, and the daily nutritional intake is calculated as body weight × 25kcal. At the same time, the consistency of food is adjusted according to the degree of swallowing disorder: DOSS-S1-2 use IDSSI4 grade thick liquid, 3-4 use IDSSI3 grade pureed food, and 5-7 use IDSSI2 grade slightly thickened liquid.

[0013] Furthermore, the acupuncture treatment assessment and adaptation module also includes a dynamic stimulation intensity adjustment unit, which adjusts the stimulation intensity using a formula. Adjust the intensity of acupuncture stimulation, where I is the real-time stimulation intensity; The initial stimulation intensity is represented by k; the adjustment coefficient is represented by t; and the acupuncture treatment time is represented by t. for The difference between the overall swallowing function score at any given time and the score before treatment. First; Integral term This indicates the cumulative change in score during treatment; when When the value is positive and the cumulative amount increases, the stimulus intensity I decreases; when When the value is negative or the cumulative amount remains unchanged, the stimulus intensity I remains unchanged or slightly increases.

[0014] Furthermore, it includes a cross-center data synchronization module, which supports data synchronization among multiple medical institutions. Synchronized content includes patient multimodal assessment data, intervention plans, and effect records. It employs a distributed data synchronization protocol, with each center having a data node. A synchronization request is triggered within 5 seconds of data changes. Before synchronization, the data hash value is calculated using the SHA-256 algorithm, and only data with inconsistent hash values ​​is synchronized. It supports offline synchronization; when offline, data is stored in a local encrypted directory and encrypted using the device's lock screen password. Synchronization is automatically initiated upon reconnection. Data synchronization permissions can be set, and center administrators can configure the synchronization scope. Synchronization logs are recorded, including synchronization time, data volume, and synchronization status.

[0015] Furthermore, it includes a user management and access control module, which categorizes users into three roles: patients, medical staff, and system administrators. Patients can only view their personal assessment reports, intervention plans, and effect records, and submit feedback. Medical staff collect data, perform assessments, configure intervention plans, and view data of patients under their jurisdiction; medical staff must log in through professional qualification certification and facial recognition. System administrators configure system parameters, manage user permissions, maintain the database, and perform model iterations. Operation logs are set up to record all user actions, and these logs are retained for at least three years. A role-based access control policy is adopted to prohibit unauthorized operations.

[0016] Compared with existing technologies, the beneficial effects of this invention are: During the assessment phase, the system deeply integrates imaging data, physiological signals, and clinical scales, enabling not only the grading of the degree of dysphagia but also the precise localization of pathological mechanisms. For example, by correlating the open cross-sectional area of ​​the UES with the FA value, it can quickly identify disorders related to dorsolateral medullary injury; and by co-analyzing pharyngeal passage time and electromyographic latency, it can clarify the type of neural drive defect, avoiding the information fragmentation caused by a single assessment tool, making the assessment results more consistent with the patient's true pathological state, and providing a precise basis for the formulation of subsequent intervention plans.

[0017] In terms of intervention adaptation, the system breaks through the limitations of traditional homogeneous solutions, enabling personalized adjustments to acupuncture and combined rehabilitation. The acupuncture treatment module precisely matches acupoints and operation parameters according to the stage of swallowing disorders, and determines the intervention intensity by combining quantitative indicators such as DOSS-S classification and PAS score. For example, for patients with severe esophageal disorders, ear acupuncture points for the esophagus and Zusanli are selected, and for patients in the swallowing stage, the insertion angle and depth of Fengchi and Renying acupoints are optimized to ensure that acupuncture stimulation directly reaches the functional impairment target and improves the treatment effect. The combined rehabilitation intervention module dynamically adjusts the concentration and duration of icteric acid stimulation and neuromuscular electrical stimulation parameters based on the patient's real-time assessment results, and combines personalized content of standard swallowing training. For example, for patients with insufficient hyoid bone displacement, laryngeal elevation training is strengthened, and for patients with high aspiration risk, food consistency and eating position are optimized, making the intervention plan more in line with the patient's functional improvement process and avoiding ineffective intervention.

[0018] Furthermore, the system significantly enhances intervention safety and long-term application value through real-time monitoring and iterative data design. The aspiration risk dynamic monitoring unit collects signals such as swallowing reflex latency and blood oxygen saturation in real time, triggering an early warning mechanism to prevent sudden risks and reduce the probability of pneumonia. Cross-center data synchronization and model iteration functions integrate clinical data from multiple institutions, continuously optimizing the machine learning fusion model, allowing assessment and intervention plans to be constantly improved with the accumulation of clinical data. Simultaneously, the nutritional support adaptation module adjusts dietary plans based on nutritional risk scores and the degree of swallowing dysphagia, ensuring patients receive adequate nutrition while reducing the risk of aspiration, thus comprehensively improving the rehabilitation efficiency and quality of life for patients with swallowing disorders. Attached Figure Description

[0019] Figure 1 This is a schematic block diagram of the integrated system for intelligent assessment and intervention of multimodal swallowing function proposed in this invention; Figure 2 A bar chart comparing the incidence of aspiration pneumonia in patients with different DOSS grades; Figure 3 A bar chart comparing the effectiveness of staged acupuncture at different swallowing stages with traditional acupuncture. Figure 4 This is a bar chart showing the weight distribution of multimodal swallowing assessment indicators. Detailed Implementation

[0020] 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.

[0021] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0023] Reference Figures 1 to 4 A multimodal intelligent assessment and intervention system for swallowing function, comprising the following modules: The multimodal data acquisition module collects multimodal data related to swallowing function, including imaging data, physiological signal data, and clinical scale data. Imaging data is acquired using a video-assisted swallowing examination device, a fiberoptic endoscopy swallowing function examination device, and a 3.0T MRI device. The video-assisted swallowing examination device uses a frame rate of at least 30fps and a contrast agent concentration of 40% w / v to acquire dynamic images of lateral swallowing of 1ml of paste-like contrast agent, recording the hyoid bone displacement trajectory and the opening process of the urethral sphincter (UES). The fiberoptic endoscopy swallowing function examination device uses a high-definition endoscope with a resolution of at least 1080P to acquire images of the pharyngeal mucosa and food residue. The 3.0T MRI device, combined with a single-shot EPI sequence and a b-value set to 1000s / mm², performs a whole-brain coronal DTI scan with a spatial resolution not exceeding 2mm³, fully covering the corticobulbar tract pathway. Physiological signal data is acquired through… Surface electromyography (EMG) and ultrasound were used to collect data. The EMG device placed electrodes on the surface projection area of ​​the thyrohyoid muscle, with a sampling frequency of 1 kHz, recording the EMG latency and amplitude. The ultrasound device used a 7.5 MHz high-frequency linear array probe, placed in the coronal plane of the anterior midline of the neck, to record the dynamic changes in the distance between the hyoid bone and thyroid cartilage in real time. Clinical scale data were entered through the system's interactive interface, including DOSS-S scale scores, osmotic-aspiration scale scores, functional oral feeding scale scores, and NRS2002 nutritional risk scores. All data were associated with a unique patient identifier and stored in DICOM, EDF, and JSON formats. The multimodal intelligent assessment module integrates and analyzes the collected multimodal data, including an image analysis unit, a physiological signal analysis unit, a scale scoring unit, and a machine learning fusion unit. The image analysis unit calculates the hyoid bone displacement amplitude using a dynamic tracking algorithm, with a threshold of 2cm to distinguish between normal and abnormal values. It also calculates the peak area of ​​the UES open cross-section using a UES open cross-section measurement algorithm, with a threshold of 85mm² to diagnose dorsolateral medullary injury-related disorders. Furthermore, it calculates the corticobulbar tract FA value using a DTI analysis algorithm, with a threshold of 0.34 to determine white matter microstructure damage. The physiological signal analysis unit calculates the cortical-laryngospheric muscle conduction latency of surface electromyography; latency exceeding 150ms is considered neurological. The system assesses dysphagia by calculating the laryngeal elevation time monitored by ultrasound; a time exceeding 300ms is considered a delayed laryngeal elevation. The scoring unit grades the DOSS-S scale as follows: 1-2 for severe dysphagia, 3-4 for moderate dysphagia, and 5-7 for mild dysphagia. The osmotic-aspiration scale is graded from 1 to 8, with ≥4 indicating high aspiration risk. The functional oral feeding scale is graded from 1 to 7, with ≤3 indicating the need for tube feeding. The machine learning fusion unit uses the XGBoost algorithm, taking image features, physiological indicators, and scale scores as input, and outputting the degree of swallowing disorder, aspiration risk probability, and nutritional risk level. Model training uses 5-fold cross-validation, with a validation set accuracy of at least 90%. The acupuncture treatment assessment and adaptation module adapts acupuncture treatment plans based on multimodal intelligent assessment results. It includes a staging assessment unit, acupoint selection unit, operation parameter configuration unit, and effect assessment unit. The staging assessment unit divides swallowing disorders into oral, swallowing, and esophageal phases. Oral phase imaging or physiological analysis shows a hyoid bone displacement amplitude <2cm or prolonged oral passage time. Swallowing phase imaging analysis shows a peak open area of ​​the UES <85mm². 2 Or, if the laryngeal elevation time is >300ms, and the corresponding imaging analysis during the esophageal phase shows an open cross-sectional area of ​​the esophageal endoscopic space (UES) ≤85mm, then... 2 Furthermore, esophageal residue may be present, or esophageal dysfunction may be suspected based on clinical assessment. The acupoint selection unit selects Lianquan, Dicang, Jiache, Jinjin, and Yuye for the oral phase to improve tongue muscle movement and oral sensation; selects Fengchi, Yamen, Tiantu, Renying, and Futu for the swallowing phase to enhance pharyngeal reflex and laryngeal elevation; and selects auricular acupuncture points such as esophagus, cardia, Zusanli, Zhongwan, and Neiguan for the esophageal phase to regulate esophageal sphincter function and gastrointestinal qi. The operation parameter configuration unit sets the needle insertion depth: 15mm for acupoints during the oral phase, 15mm for Fengchi (GB20) during the swallowing phase, 15mm for Renying (ST9) obliquely towards the larynx, and 30mm for Futu (ST1) obliquely towards the larynx during the swallowing phase, and 20mm for acupoints during the esophageal phase. The needle retention time is 30 minutes for all acupoints. During the oral phase, Jinjin (BL51) and Yuye (BL62) are needled 3-5 times, and the posterior pharyngeal wall is needled with long needles during the swallowing phase. The effect evaluation unit evaluates the acupuncture effect through the Kubota water swallowing test and the standard swallowing function evaluation scale. A Kubota water swallowing test score of 1-2 is considered effective, and a reduction of ≥5 points in the standard swallowing function evaluation scale score is considered effective. The combined rehabilitation intervention module configures a combined rehabilitation plan based on assessment results and acupuncture effects. It includes a citric acid stimulation unit, a standard swallowing training unit, and a neuromuscular electrical stimulation unit. The citric acid stimulation unit uses a citric acid concentration gradient: 4% in week 1 and 8% in week 2, with a stimulation temperature of 4°C. It is applied to the base of the tongue and the posterior pharyngeal wall, with each stimulation session lasting ≤5 minutes, twice daily with an interval of ≥6 hours. The standard swallowing training unit includes oral motor training, pharyngeal stimulation training, and feeding adjustment training. Oral motor training includes tongue protrusion and retraction, and lateral movement, 3 sets per day. 10 sessions of resistance training, including gently pressing the back of the tongue with a tongue depressor while the patient pushes it upwards, and lip-pursing and blowing exercises, 2 sets of 5 repetitions each day; pharyngeal stimulation training, including gently touching the palatal arch and posterior pharyngeal wall with an ice cotton swab for 5 minutes each time, supraglottic swallowing and Mendelssohn's maneuver; food adjustment training, selecting the consistency of food according to the IDSSI level, sitting with the body leaning forward and the chin slightly tucked, with each mouthful ≤5mL; the neuromuscular electrical stimulation unit uses NMES technology, with electrodes placed under the chin and neck, frequency set at 50Hz, pulse width set at 300μs, and current intensity adjusted to the patient's tolerance, 20 minutes each time, once a day; The results feedback and system iteration module generates multimodal assessment and intervention reports, which include basic patient information, multimodal data characteristics, assessment results, acupuncture plans, combined rehabilitation plans, and intervention effects. It records assessment data after each intervention, including changes in imaging features, physiological indicators, and scale scores, stored in a distributed database using AES256 encryption for transmission and storage, with daily full backups and real-time incremental backups. The module iterates the machine learning fusion model using newly added assessment and intervention data, retraining the model quarterly. Emergency iteration is triggered when the model's prediction accuracy drops by more than 5%, and the validation set accuracy must recover to over 90% after iteration. Simultaneously, the module optimizes acupuncture point selection and combined rehabilitation plan parameters based on the intervention effects.

[0024] This invention also includes a comprehensive swallowing function scoring module, which calculates the swallowing function score using a formula. Calculate the patient's overall swallowing function score, where S represents the overall swallowing function score, ranging from 0 to 100. As image evaluation weight, Weighting of physiological signals For the scale scores, and , Values ​​0.4 Values ​​0.3 The value is 0.3, determined by a machine learning model trained on data from 3000 historical patients; S1 is the imaging assessment score, ranging from 0 to 100, obtained by summing the hyoid bone displacement score, UES open cross-sectional area score, FA value score, and pharyngeal passage time score. Hyoid bone displacement ≥2cm receives 25 points, 1-2cm receives 15 points, and <1cm receives 5 points; UES open cross-sectional area ≥100mm² receives 25 points, 85-100mm² receives 100mm², and 85-100mm² receives 100mm². =15 points for ², 5 points for <85mm², 25 points for FA value ≥0.4, 15 points for 0.34-0.4, and 5 points for <0.34, 25 points for pharyngeal passage time <850ms, 15 points for 850-1000ms, and 5 points for >1000ms; S2 is the physiological signal assessment score, ranging from 0 to 100 points, obtained by summing the electromyographic latency score and the laryngeal elevation time score, with an electromyographic latency <150ms scoring 50 points. The time for laryngeal elevation is calculated as follows: 150-200ms = 30 points, >200ms = 10 points; <300ms = 50 points, 300-400ms = 30 points, >400ms = 10 points. S3 is the scale score, ranging from 0 to 100 points, obtained by summing the DOSS-S score, PAS score, FOIS score, and NRS2002 score. A DOSS-S score of 7 is worth 30 points, 5-6 is worth 20 points, and 3-4 is worth 10 points. A score of 10 is awarded for a PAS grade of 1-2, 5 for 1-2, 30 for 3-4, 20 for 5-6, and 5 for 7-8. A score of 7 for FOIS is awarded for FOIS, 20 for FOIS, 15 for FOIS, 10 for FOIS, and 5 for FOIS. A score of 20 for NRS200 is awarded for NRS200, 15 for FOIS, and 5 for FOIS. This score can directly reflect the patient's overall swallowing function level and provide a quantitative basis for the adaptation of intervention plans.

[0025] This invention also includes a multimodal data quality verification module, which verifies the quality of acquired image data and physiological signal data. Image data verification indicators include the clarity of video-assisted swallowing examination images, the mucosal visibility of fiberoptic endoscopy swallowing function examination images, and the signal-to-noise ratio of MRI DTI images. For video-assisted swallowing examination images, the intra-frame contrast agent boundary clarity is required to be ≥90%, calculated using an edge detection algorithm; if the clarity is below this value, a prompt is made to re-acquire and adjust the device focus. For fiberoptic endoscopy swallowing function examination mucosal visibility, the mucosal texture recognition rate is required to be ≥85%, calculated using a texture feature extraction algorithm; if the recognition rate is below this value, a prompt is made to clean the endoscope lens. MRI... The DTI image signal-to-noise ratio (SNR) must be ≥20dB, calculated by the ratio of the standard deviation of signal intensity to the standard deviation of noise. If the SNR is lower than this value, the scanning time should be extended. Physiological signal data verification indicators include surface electromyography (SEMG) SNR and ultrasound image resolution. The surface EMG SNR must be ≥30dB, calculated by the ratio of EMG signal amplitude to noise amplitude. If the SNR is lower than this value, the electrode position and skin impedance should be adjusted. The ultrasound image resolution must have a hyoid-thyroid cartilage structure resolution of ≥95%, calculated by a structural edge extraction algorithm. If the resolution is lower than this value, the probe position and frequency should be adjusted. Data that passes the verification enters the evaluation module, while data that fails the verification is marked with the reason and guided to be re-acquired.

[0026] In this invention, the acupuncture treatment assessment and adaptation module further includes an acupoint positioning and navigation unit. This unit uses ultrasound guidance to locate acupoints. During the oral cavity phase, the ultrasound probe is placed in the anterior midline of the neck to display the soft tissue layer below the hyoid bone corresponding to the Lianquan acupoint, marking the needle insertion point and depth. During the swallowing phase, the ultrasound probe is placed below the occipital bone in the back of the neck to display the soft tissue lateral to the foramen magnum corresponding to the Fengchi acupoint, determining the needle insertion direction and angle. During the esophageal phase, the high-frequency ultrasound probe is placed in the concha cavity to display the junction of cartilage and soft tissue corresponding to the esophageal acupoint, marking the puncture point. Simultaneously, the ultrasound image during the needle insertion process is displayed in real time to avoid damage to blood vessels and nerves. The positioning error is controlled within ±1mm, improving the accuracy and safety of acupoint positioning.

[0027] This invention also includes a rehabilitation effect prediction module, which uses a formula... The probability of effective rehabilitation intervention is predicted, where P is the probability of effective intervention, ranging from 0 to 1, with a value closer to 1 indicating a higher likelihood of effective intervention; t is the duration of intervention, in days. The difference between the comprehensive swallowing function score before and during the intervention. =Smiddle-Sbefore, with a value range of -20 to 50 points; a is the time coefficient, with a value of 0.02, obtained by fitting historical data, reflecting the impact of intervention time on the effect; b is the score change coefficient, with a value of 0.05, reflecting the impact of score improvement on the effect; c is a constant term, with a value of -1.2, determined through training with intervention data from 300 patients; when P ≥ 0.8, the intervention plan is deemed suitable, and the current plan continues; when 0.5 ≤ P < 0.8, the intervention parameters are adjusted, such as increasing the frequency of ice acid stimulation or extending the acupuncture needle retention time; when P < 0.5, the intervention plan is reassessed and changed, such as switching from standard swallowing training to neuromuscular electrical stimulation combined with acupuncture. This prediction can determine the intervention effect in advance and avoid ineffective intervention.

[0028] In this invention, the combined rehabilitation intervention module also includes a dynamic monitoring unit for aspiration risk. This unit collects physiological signals from the patient's swallowing process in real time, including the swallowing reflex latency from surface electromyography, the laryngeal elevation delay time from ultrasound monitoring, and blood oxygen saturation. Monitoring thresholds are set; when the swallowing reflex latency exceeds 200ms, the laryngeal elevation delay time exceeds 400ms, or the blood oxygen saturation drops below 94%, a real-time warning is triggered, notifying medical staff through system interface prompts and audible alarms. Simultaneously, the swallowing scenario at the time of the warning is recorded, including food consistency, eating position, and number of swallows. The cause of the warning is analyzed; for example, if the food consistency is too low, it is recommended to increase the consistency; if the eating position is inappropriate, it is recommended to adjust to a mandibular retraction position. This monitoring can prevent aspiration risk in real time and improve the safety of the intervention process.

[0029] This invention also includes a nutritional support adaptation module, which configures a nutritional plan based on the NRS2002 nutritional risk score and serum prealbumin level. A high nutritional risk is identified when the NRS2002 score is ≥3 or serum prealbumin is <180mg / L, and a high-energy-density formulation is configured with an energy density of 1.5kcal / ml, with daily nutritional intake calculated as body weight × 35kcal. A moderate nutritional risk is identified when the NRS2002 score is 1-2 and serum prealbumin is 180-220mg / L, and a conventional-energy-density formulation is configured with an energy density of [missing information]. 1.0 kcal / ml, daily nutrient intake is calculated as body weight × 30 kcal; when the NRS2002 score is 0 and serum prealbumin ≥ 220 mg / L, it is judged as low nutritional risk, and a normal diet plan is formulated, with daily nutrient intake calculated as body weight × 25 kcal; at the same time, the consistency of food is adjusted according to the degree of swallowing dysphagia: DOSS-S1-2 use IDSSI4 grade thick liquid, 3-4 use IDSSI3 grade pureed food, and 5-7 use IDSSI2 grade slightly thickened liquid, to ensure nutrient intake while reducing the risk of aspiration.

[0030] In this invention, the acupuncture treatment assessment and adaptation module further includes a dynamic stimulation intensity adjustment unit, which adjusts the stimulation intensity using a formula. Adjust the intensity of acupuncture stimulation, where I is the real-time stimulation intensity in mA; The initial stimulation intensity was set at 1 mA for the oral phase, 1.5 mA for the swallowing phase, and 2 mA for the esophageal phase; k was the adjustment coefficient, with a value of 0.01, obtained by fitting acupuncture data from 50 patients; t was the acupuncture treatment time in minutes. for The difference between the overall swallowing function score at any given time and the score before treatment. First; Integral term This indicates the cumulative change in score during treatment, expressed in minutes and minutes; when When the value is positive and the cumulative amount increases, the stimulus intensity I decreases to avoid overstimulation; when When the value is negative or the cumulative amount remains unchanged, the stimulation intensity I is maintained or slightly increased to ensure the stimulation effect. This adjustment can optimize the stimulation intensity according to the patient's real-time response and improve the suitability of acupuncture treatment.

[0031] This invention also includes a cross-center data synchronization module, which supports data synchronization among multiple medical institutions. Synchronized content includes patient multimodal assessment data, intervention plans, and effect records. It employs a distributed data synchronization protocol, with each center setting up a data node. A synchronization request is triggered within 5 seconds of data changes. Before synchronization, a data hash value is calculated using the SHA-256 algorithm, and only data with inconsistent hash values ​​is synchronized. It supports offline synchronization; when offline, data is stored in a local encrypted directory and encrypted using a device lock screen password. Synchronization is automatically initiated upon reconnection. Data synchronization permissions can be set, allowing center administrators to configure the synchronization scope, such as synchronizing only anonymized assessment results and prohibiting the synchronization of patient privacy data. Synchronization logs are recorded, including synchronization time, data volume, and synchronization status, facilitating traceability of the synchronization process. This module enables multi-center data sharing, providing more data support for model iteration.

[0032] This invention also includes a user management and access control module, which categorizes users into three roles: patients, medical staff, and system administrators. Patients can only view their personal assessment reports, intervention plans, and effect records, and submit feedback. Medical staff can collect data, perform assessments, configure intervention plans, and view data of patients under their jurisdiction; medical staff must log in through professional qualification certification and facial recognition. System administrators can configure system parameters, manage user permissions, maintain the database, and perform model iterations. An operation log is set up to record all user actions, including data collection, assessment execution, plan configuration, and data viewing; the log is retained for at least three years. A role-based access control policy is adopted to prohibit unauthorized operations, ensuring the security of system data access and the traceability of operations.

[0033] The following two examples further illustrate the specific implementation of this system: Example 1: Application scenario for patients with dysphagia after acute ischemic stroke This embodiment focuses on patients with dysphagia following acute ischemic stroke within 72 hours of onset, including 60 patients (35 males and 25 females, aged 50-82 years, with a mean age of 65.4 years). All patients were diagnosed with ischemic stroke by CT or MRI-DWI and assessed as having dysphagia grades 1-7 on the DOSS-S scale, with 18 cases of severe, 22 cases of moderate, and 20 cases of mild dysphagia. This embodiment emphasizes the integration of acupuncture treatment with icteric acid stimulation as a combined intervention to verify the accuracy and effectiveness of the system in assessing dysphagia in post-stroke patients. The system's designed comprehensive swallowing function scoring formula, rehabilitation effect prediction formula, and acupuncture stimulation intensity formula were applied.

[0034] 1. Detailed Implementation of Core Modules The multimodal data acquisition module employs a multi-device collaborative approach. Image data is acquired via CT scanner, 3.0T MRI scanner, and digital gastrointestinal imaging system. CT scans are used to rule out cerebral hemorrhage, and the 3.0T MRI scanner uses DTI sequences with a b-value set to 1000 s / mm. 2 It employs 32-directional scanning, with spatial resolution controlled at 2mm. 3 The digital gastrointestinal machine was used for VFSS examination at a frame rate of 30fps, with 40% w / v barium sulfate paste as the contrast agent. Physiological signal data were acquired using a surface electromyography (SEMG) device and an EPIQ5 ultrasound device. The electrodes of the SEMG device were placed on the surface projection area of ​​the thyrohyoid muscle, with a sampling frequency of 1kHz and a bandwidth range of 20-500Hz. The ultrasound device used a 7.5MHz high-frequency linear array probe, placed in the coronal plane of the anterior midline of the neck, with a scanning depth of 3-5cm. Clinical scale data were entered through the system touch screen, including the DOSS-S scale, PAS scale, FOIS scale, and NRS2002 scale. The DOSS-S scale was scored independently by two rehabilitation physicians and the average value was taken. The PAS scale was based on VFSS image scores. The FOIS scale was combined with dietary observation. The NRS2002 scale was combined with serum prealbumin detection. Serum prealbumin detection was performed using a fully automated biochemical analyzer with a detection limit of 10mg / L. All data is linked to the patient's hospitalization ID. Imaging data is stored in DICOM format, physiological signals are stored in EDF format, and scale data is stored in JSON format. All data is stored uniformly in the hospital's distributed storage system, which uses the HDFS architecture.

[0035] The multimodal intelligent assessment module's image analysis unit implemented the algorithm using MATLAB R2023a. Hyoid bone displacement dynamic tracking employed optical flow to extract hyoid bone coordinates and calculate displacement amplitude. With a normal threshold set at 2cm, the average displacement was 1.2cm in 18 severe patients, 1.8cm in 22 moderate patients, and 2.3cm in 20 mild patients. The open cross-sectional area of ​​the UES was measured frame-by-frame using VFSS images, with the closure point between the posterior wall of the cricoid cartilage and the posterior pharyngeal wall as the boundary. The average open area for severe patients was 78mm². 2 The average open area for moderate patients was 95 mm. 2 The average open area for mild cases is 120 mm. 2 DTI analysis used FSL software to calculate the FA value of the corticobulbar tract. The average FA value was 0.28 for severe patients, 0.32 for moderate patients, and 0.38 for mild patients. The healthy reference value was 0.49 ± 0.05. The physiological signal analysis unit calculated the electromyographic latency and laryngeal elevation time. The average electromyographic latency was 180 ms for severe patients, 160 ms for moderate patients, and 140 ms for mild patients, with an abnormal threshold of 150 ms. The average laryngeal elevation time was 350 ms for severe patients, 320 ms for moderate patients, and 280 ms for mild patients, with an abnormal threshold of 300 ms. The scale scoring units were based on the DOSS-S grading standard, with grades 1-2 indicating severe, 3-4 indicating moderate, and 5-7 indicating mild. In the PAS score, severe patients averaged 5.2, moderate patients averaged 3.8, and mild patients averaged 2.1, with the high aspiration risk threshold set at ≥4. In the FOIS score, severe patients averaged 2.1, moderate patients averaged 3.5, and mild patients averaged 5.2, with the tube feeding support threshold set at ≤3. The machine learning fusion unit uses the Python XGBoost library. It takes 200 historical patient data as input, including 60 patient data from this embodiment. Features include hyoid bone displacement, UES area, FA value, electromyographic latency, and DOSS-S score. Five-fold cross-validation is used, with a validation accuracy of 92%. The output includes patient classification, aspiration risk probability, and nutritional risk level. Finally, it is determined that there are 18 severe cases, 22 moderate cases, and 20 mild cases. The aspiration risk probability is 0.78 for severe patients, 0.55 for moderate patients, and 0.22 for mild patients. The nutritional risk level is 12 high risk, 28 moderate, and 20 low risk.

[0036] The staging evaluation unit of the acupuncture treatment evaluation and adaptation module classifies patients into oral phase, pharyngeal phase, and esophageal phase based on multimodal evaluation indicators. Among them, patients in the oral phase (20 cases) showed that image analysis showed a hyoid bone displacement amplitude < 2 cm or an extended oral passage time; patients in the pharyngeal phase (22 cases) showed that image analysis showed a UES opening peak area < 85 mm² or a laryngeal elevation time > 300 ms; patients in the esophageal phase (18 cases) showed that image analysis showed a UES opening cross-sectional area ≤ 85 mm² and accompanied by esophageal phase residue. The acupoint selection unit determines acupoints according to the staging of the above-mentioned dysfunction sites. For the oral phase, Lianquan, Dicang, Jiache, Jinjin, and Yuye are selected. The location of the Lianquan acupoint is 1 cm below the hyoid bone, the location of the Dicang acupoint is 0.4 cun outside the corner of the mouth, the location of the Jiache acupoint is 1 horizontal finger above the anterior superior angle of the mandibular angle, the location of the Jinjin acupoint is on the left side of the frenulum linguae, and the location of the Yuye acupoint is on the right side of the frenulum linguae; for the swallowing phase, Fengchi, Yamen, Tian突, Renying, and Futu are selected. The location of the Fengchi acupoint is between the sternocleidomastoid muscle and the trapezius muscle under the occipital bone, the location of the Yamen acupoint is 0.5 cun directly above the midpoint of the posterior hairline, the location of the Tian突 acupoint is in the center of the suprasternal fossa, the location of the Renying acupoint is 1.5 cun beside the laryngeal prominence, and the location of the Futu acupoint is 1.5 cun behind the Renying acupoint; for the esophageal phase, auricular acupuncture points of the esophagus, cardia, Zusanli, Zhongwan, and Neiguan are selected. The location of the auricular acupuncture point of the esophagus is in the middle of the concha of the ear, the location of the auricular acupuncture point of the cardia is in the lower part of the concha of the ear, the location of the Zusanli acupoint is 3 cun below Dubi, the location of the Zhongwan acupoint is 4 cun above the umbilicus, and the location of the Neiguan acupoint is 2 cun above the transverse wrist crease. The operation parameter configuration unit determines the specifications of the filiform needles and operation details. For the acupoints in the oral phase, filiform needles of 0.30 mm × 40 mm are used, the insertion depth is 15 mm, and the needles are retained for 30 minutes after obtaining qi. For Jinjin and Yuye, filiform needles of 0.25 mm × 25 mm are used for pricking 3 - 5 times, and 0.1 - 0.2 ml of blood is bled after pricking; for the swallowing phase, the Fengchi acupoint is inserted 15 mm towards the tip of the nose, the Renying acupoint is obliquely inserted 15 mm towards the larynx, the Futu acupoint is obliquely inserted 30 mm towards the larynx, and after obtaining qi, the high-frequency twirling reinforcing method is used for 1 minute and the needles are retained for 30 minutes; for the esophageal phase, auricular acupuncture uses filiform needles of 0.20 mm × 13 mm, the insertion depth is 2 mm, and the needles are retained for 30 minutes, and the insertion depth of the body acupoints is 20 mm and the needles are retained for 30 minutes. The effect evaluation unit conducts an evaluation 4 weeks after treatment, using the Wada drinking water test and the SSA scale. Grade 1 of the Wada drinking water test is considered markedly effective, grade 2 is considered effective, and a decrease in the SSA scale score of ≥ 5 points is considered effective. Finally, among the 20 patients in the oral phase, 15 were markedly effective and 3 were effective; among the 22 patients in the swallowing phase, 12 were markedly effective and 8 were effective; among the 18 patients in the esophageal phase, 5 were markedly effective and 10 were effective.

[0037] The combined rehabilitation intervention module's icy acid stimulation unit was configured with citric acid concentration and operating parameters. Analytical grade citric acid was used to prepare the concentration. In the first week, the concentration was 4% and the pH was 2.5. In the second week, the concentration was 8% and the pH was 2.0. After preparation, it was refrigerated at 4°C. When using, 0.1 ml was applied with a sterile cotton swab and gently wiped on the patient's tongue base and posterior pharyngeal wall. The stimulation time was 4 minutes per session to avoid mucosal damage. Stimulation was performed twice a day, with an 8-hour interval between the two stimulations. 18 severe patients used the 4% concentration for the first 2 weeks and the 8% concentration for the next 2 weeks. The remaining patients used the 4% concentration throughout the entire course. The standard swallowing training unit includes three types of training content: oral motor training, including tongue protrusion and retraction, and left and right swinging, 3 sets of 10 repetitions each day; resistance training, using a tongue depressor to gently press the back of the tongue while the patient pushes it upward, 2 sets of 8 repetitions each day; lip-pursing and blowing exercises, 2 sets of 5 repetitions each day; and puffing out the cheeks, holding for 5 seconds, and then slowly exhaling; pharyngeal stimulation training, including gently touching the palatal arch and posterior pharyngeal wall with an ice swab for 5 minutes each time, once a day; supraglottic swallowing method, following the deep inhalation-breath-swallowing-coughing-re-swallowing sequence, 3 sets of 5 repetitions each day; and Mendel's phonation. The relaxation technique involves gently pushing the thyroid cartilage upwards with your fingers during swallowing and holding for 2 seconds, 2 sets of 5 repetitions each day. Feeding adjustment training determines the food consistency and feeding method based on the stage of the disease. Patients in the oral stage use IDDSI grade 2 slightly thickened liquids; patients in the swallowing stage use IDDSI grade 3 pureed foods; and patients in the esophageal stage use IDDSI grade 4 thick liquids. Patients sit with their bodies leaning forward 30° and their chin slightly tucked in. Each feeding is 3-5 mL, and the patient is only fed the next bite after confirming swallowing. The patient remains seated for 30 minutes after each feeding. The neuromuscular electrical stimulation unit was used only in 18 severely affected patients. A Vitalstim 5900 electrical stimulator was used, with electrodes placed under the chin and neck, the cathode under the chin and the anode at the upper edge of the thyroid cartilage. The frequency was set to 50 Hz, the pulse width to 300 μs, and the current intensity increased from 1 mA to the patient's tolerance, with an average current intensity of 3 mA. Each stimulation session lasted 20 minutes and was performed once daily.

[0038] The formula for comprehensive swallowing function assessment is based on the formula. S represents the comprehensive swallowing function score, ranging from 0 to 100 points; The image evaluation weight is set to 0.4. The weight for evaluating physiological signals is set to 0.3. The scale score weight is 0.3, and S1 is the imaging assessment score, ranging from 0 to 100; S2 is the physiological signal assessment score, ranging from 0 to 100; S3 is the scale score, ranging from 0 to 100. Taking a severe patient as an example, in the S1 calculation, hyoid bone displacement of 1.0 cm scored 5 points, UES open area of ​​75 mm² scored 5 points, FA value of 0.28 scored 5 points, and pharyngeal passage time of 900 ms scored 5 points, so S1 = 20 points. In the S2 calculation, electromyographic latency of 180 ms scored 10 points, and laryngeal elevation time of 350 ms scored 10 points, so S2 = 20 points. In the S3 calculation, DOSS-S2 grade scored 5 points, PAS6 grade scored 5 points, FOIS2 grade scored 5 points, and NRS20023 score scored 5 points, so S3 = 20 points. The final S = 0.4 × 20 + 0.3 × 20 + 0.3 × 20 = 20 points. The lower the score, the worse the swallowing function. The rehabilitation effect is predicted using the formula. Where P is the probability of effective intervention, ranging from 0 to 1, with a value closer to 1 indicating a higher likelihood of effective intervention; a is the time coefficient, with a value of 0.02; t is the duration of intervention, in days; b is the coefficient of score change, with a value of 0.05; ΔS is the difference between the comprehensive swallowing function score before and during intervention, ΔS = S_intermediate - S_previously, ranging from -20 to 50 points; c is a constant term, with a value of -1.2. After 2 weeks of treatment, for this severe patient, t = 14 days, ΔS = S_intermediate - S_previously = 35 - 20 = 15 points. Substituting into the formula, we get P = 1 / (1 + exp(-(0.02 × 14 + 0.05 × 15 - 1.2))) = 1 / (1 + exp(-(0.28 + 0.75 - 1.2))) = 1 / (1 + exp(-0.03)) ≈ 0.507. It is determined that the intervention plan needs to be adjusted, increasing the frequency of ice acid stimulation to 3 times per day.

[0039] 2. Effect Verification and Table Analysis Table 1: Comparison of key assessment indicators before and after intervention in patients in Example 1 (covering 60 patients)

[0040] Table 1 shows the data based on a 4-week intervention in 60 patients with dysphagia following acute ischemic stroke, fully demonstrating the advantages of the system in this scenario. Before the intervention, the patients' overall swallowing function was poor, with a DOSS-S score of only 3.2 and a FOIS score of 3.1. 18.3% of the patients had already developed aspiration pneumonia, and low serum prealbumin levels indicated nutritional risks. These indicators reflected the dual predicament of patients' swallowing function and nutritional status. After the intervention, the DOSS-S and FOIS scores significantly improved, indicating that the patients' swallowing function and independent feeding ability were effectively improved; the PAS score decreased significantly, indicating a substantial reduction in the risk of aspiration; the incidence of aspiration pneumonia decreased to 5.0%, which is attributed to the synergistic effect of the system in enhancing gag reflex sensitivity through ice acid stimulation, improving nerve drive function through acupuncture, and real-time monitoring and early warning of aspiration risk, effectively reducing the occurrence of complications; the serum prealbumin level increased because the nutritional support module accurately configured high-energy-density preparations based on the NRS2002 score, meeting the patients' nutritional needs while ensuring swallowing safety. Compared with traditional single intervention methods, this system accurately locates the type of disorder through multimodal assessment and integrates multiple intervention methods such as acupuncture, ice acid stimulation, and standard training to achieve simultaneous improvement in swallowing function and nutritional status. The effect is more comprehensive and stable. Especially for severe patients, the FOIS score can be improved by more than 2.0 points, which further verifies the clinical value of the integrated design.

[0041] Example 2: Application scenario for patients with dysphagia due to dorsolateral medullary injury This embodiment focuses on patients with dysphagia caused by dorsolateral medullary injury, including 50 patients (28 males and 22 females), aged 55-85 years (mean age 68.2 years). All patients were diagnosed with dorsolateral medullary lesions by MRI-DWI, located in the retroollioid region or nucleus ambiguus, with DOSS-S scores of 1-4, including 20 severe cases and 30 moderate cases. This embodiment emphasizes combined intervention with imaging assessment and neuromuscular electrical stimulation. The imaging assessment integrates DTI and VFSS techniques, and references relevant research and multimodal assessment technology documents on dorsolateral medullary injury, applying a systematically designed acupuncture stimulation intensity formula.

[0042] 1. Detailed Implementation of Core Modules The multimodal data acquisition module acquired image data using three types of equipment: a VF device with a lateral camera frame rate of 30fps, where the patient was in an upright lateral position with a cervical flexion angle of 60°, swallowing 1ml of 40% w / v barium paste, and the hyoid bone displacement trajectory and UES opening process were recorded throughout; a 3.0T MRI used a single-shot EPI sequence with a b-value of 1000s / mm², employing 32-direction coronal DTI scanning with a spatial resolution of 1.8mm³, and the scanning range completely covered the corticobulbar tract pathway, extending from the corona radiata to the posterior limb of the internal capsule and then to the medulla oblongata; and a FEES with a resolution of 1080P was used to observe the pharyngeal mucosa condition and food residue. Physiological signal data were acquired using two types of equipment. Surface electromyography (EMG) was performed using a Keypoint.net device, with electrodes placed in the thyrohyoid muscle surface projection area, located between the upper edge of the thyroid cartilage and the hyoid bone. The sampling frequency was 2 kHz, recording the EMG latency and amplitude. Ultrasound was performed using a Logiq E9 device equipped with a 7.5 MHz high-frequency linear array probe, scanning the hyoid-thyroid cartilage distance in the coronal plane, with a sampling frequency of 25 fps. Clinical scale data were entered through the system interface, including the DOSS-S scale, PAS scale, NRS2002 scale, and FA value. The DOSS-S scale was independently scored by two neurologists. The PAS scale was based on VFSS image scoring. The NRS2002 scale was combined with serum prealbumin detection using a Roche cobas 8000 detection device. The FA value was calculated using FSL software, with the region of interest set to the middle and posterior third of the posterior limb of the internal capsule. Data storage follows standardized formats: image data is stored in DICOM format, physiological signal data is stored in EDF format, and scale data is stored in JSON format. All data is associated with a unique patient ID and stored in the hospital's PACS system after encryption.

[0043] The multimodal intelligent assessment module's image analysis unit uses specialized software for quantitative analysis. Hyoid bone displacement was measured using ImageJ software, with an average displacement of 1.5 cm for severe patients and 1.9 cm for moderate patients. The open cross-sectional area of ​​the UES was measured frame-by-frame using VFSS images, with the closure point between the posterior wall of the cricoid cartilage and the posterior pharyngeal wall as the boundary. The average open area was 80 mm² for severe patients and 98 mm² for moderate patients. The DTI analysis used FSL software to calculate the FA value of the corticobulbar tract. The average FA value was 0.30 for severe patients and 0.34 for moderate patients. The healthy reference value was 0.49 ± 0.05. FA ≤ 0.34 was set as the criterion for judging white matter microstructure damage. The physiological signal analysis unit calculated key indicators: electromyographic latency averaged 190ms for severe patients and 170ms for moderate patients; laryngeal elevation time averaged 380ms for severe patients and 330ms for moderate patients; UES opening-to-laryngeal closure time averaged 65ms for severe patients and 45ms for moderate patients, with an abnormal threshold of 50ms. The scale scoring unit performed grading according to standards: in the DOSS-S grade, 20 patients were classified as severe (grades 1-2) and 30 patients as moderate (grades 3-4); the average PAS score for moderate to severe patients was 5.8 and for moderate patients it was 4.2; the average FOIS score for moderate to severe patients was 1.8 and for moderate patients it was 3.2; and in the NRS2002 score, 18 patients with a score ≥3 were considered high-risk and 12 patients with a score ≥3 were considered high-risk. The machine learning fusion unit uses the XGBoost model. Input features include FA value, UES area, electromyographic latency, UES opening-laryngeal closing time difference, and DOSS-S score. The training data includes data from 150 patients with medullary injury. Five-fold cross-validation was used, with a validation accuracy of 93%. The output results are: 20 cases of severe injury with an aspiration risk of 0.82, 30 cases of moderate injury with an aspiration risk of 0.58, and 20 cases of high nutritional risk, 22 cases of moderate risk, and 8 cases of low risk.

[0044] The acupuncture treatment and combined rehabilitation module included acupuncture treatment for 30 patients with moderate symptoms, divided into the swallowing stage. Selected acupoints included Fengchi (GB20), Yamen (GV15), Renying (ST9), and the posterior pharyngeal wall. Fengchi was needled 15mm towards the tip of the nose, Yamen was needled perpendicularly 8mm, and Renying was needled obliquely 15mm towards the larynx. The posterior pharyngeal wall was needled 3 times with long needles. All needles were 0.30mm×40mm filiform needles. After obtaining the Qi sensation, the needles were retained for 30 minutes. Treatment was given once a day. The combined rehabilitation program included three intervention methods. Neuromuscular electrical stimulation was used on 20 severely ill patients using a domestically produced KT-90A electrical stimulator. Electrodes were placed under the chin and in the suprasternal notch, with the cathode under the chin and the anode in the suprasternal notch. The frequency was set to 50Hz, the pulse width to 300μs, and the initial current to 1mA, gradually increasing to the patient's tolerance, with an average current intensity of 3.5mA. Each stimulation session lasted 20 minutes, once daily. Standard swallowing training covered all patients. Oral motor training primarily consisted of tongue resistance training, with 3 sets of 10 repetitions each daily, including pharyngeal... Training was conducted using the Mendelssohn maneuver, with two sets of five sessions per day. Feeding was adjusted using IDDSI grade 4 thick liquid, with each feeding consisting of 3 mL. Patients were positioned on their unaffected side. Nutritional support was tailored to risk level: 20 high-risk patients received a 1.5 kcal / ml energy supplement, totaling 1500 kcal per day; 22 moderate-risk patients received a 1.0 kcal / ml energy supplement, totaling 1200 kcal per day; and 8 low-risk patients followed a normal diet, totaling 1000 kcal per day.

[0045] The acupuncture stimulation intensity is adjusted using a formula. Where I is the real-time stimulus intensity, in mA; The initial stimulation intensity is set to 1.5 mA for patients in the swallowing phase; k is the adjustment coefficient, with a value of 0.01; t is the acupuncture treatment time, in minutes. for The difference between the overall swallowing function score at any given time and the score before treatment. First; Integral term This represents the cumulative change in score during treatment, expressed in points per minute. For example, in a moderate patient, after 10 minutes of treatment... =S(10)-S_front=45-30=15 points, integral term Substituting the values ​​into the formula, we get I = 1.5 × exp(-0.01 × 150) = 1.5 × exp(-1.5) ≈ 1.5 × 0.223 = 0.335 mA. By reducing the stimulation intensity, we can avoid over-excitation of nerve tissue and ensure the safety of treatment.

[0046] 2. Effect Verification and Table Analysis Table 2: Correlation between FA value and UES open area and intervention effect in patients in Example 2 (covering 50 patients)

[0047] Table 2 shows the data based on a 6-week intervention in 50 patients with dysphagia due to dorsolateral medullary injury, highlighting the system's application value in neurogenic dysphagia. Before the intervention, the patients' FA value was only 0.32, lower than the healthy reference value, and the open area of ​​the UES was 88.5 mm. 2 The findings suggest white matter microstructure damage and impaired coordination of the cricopharyngeal muscle, with a 0% recovery rate for normal eating, reflecting the severe impact of medullary injury on swallowing function. After intervention, the FA value increased to 0.38, and the open area of ​​the UES increased to 115.3 mm. 2 Furthermore, the FA value showed a strong positive correlation with the open area of ​​the UES (Ultra-Upper Esophageal Spectrum), with a correlation coefficient of 0.81. This indicates that the system accurately locates the white matter lesion area through DTI and uses targeted neuromuscular electrical stimulation to promote neural remodeling, effectively improving the opening function of the cricopharyngeal muscle. The electromyographic latency was shortened from 180.2 ms to 150.5 ms, and the UES-larynx closure time difference was shortened from 55.3 ms to 38.6 ms, suggesting a significant improvement in neural drive efficiency and temporal coordination. The normal eating recovery rate reached 56%, with a recovery rate of 73.3% for moderate patients and 25% for severe patients. The difference stems from the system's differentiated treatment plan based on the severity of the patient's injury. For moderate patients, acupuncture was emphasized to improve nerve conduction function, while for severe patients, neuromuscular electrical stimulation and nutritional support were strengthened. Compared with traditional intervention programs that rely solely on VFSS assessment, this system integrates DTI imaging and physiological signals to achieve a dual-dimensional assessment of "structure and function," more accurately targeting the pathological mechanism and resulting in more significant intervention effects. It provides a scientifically feasible solution for complex swallowing disorders such as dorsolateral medullary lesions.

[0048] refer to Figure 2 This chart visually demonstrates the effectiveness of DOSS-based intervention in controlling aspiration risk. Traditional uniform interventions, failing to differentiate between severity levels of dysphagia and employing the same care protocols, resulted in a pneumonia incidence rate as high as 15.0%. In contrast, this invention develops differentiated protocols based on DOSS scores. Patients with severe dysphagia (levels 1-2) received intermittent oral feeding combined with neuromuscular electrical stimulation for targeted aspiration prevention, reducing the pneumonia incidence rate to 6.5%. Patients with mild dysphagia (levels 5-7) underwent compensatory strategy training, resulting in a pneumonia incidence rate of only 1.0%. The overall incidence rate was 4.1%, significantly lower than traditional methods. This gradient decline trend confirms the scientific validity of matching graded interventions with dysphagia severity, effectively reducing the risk of complications in patients with varying degrees of dysphagia.

[0049] refer to Figure 3The chart clearly demonstrates the targeted advantages of staged acupuncture. Traditional acupuncture uses fixed combinations of acupoints, failing to consider the pathological differences at different stages of swallowing. For example, the oral stage focuses on tongue muscle movement disorders, the swallowing stage on weakened gag reflexes, and the esophageal stage on sphincter coordination problems, resulting in an effectiveness rate of less than 70% for each stage. In contrast, staged acupuncture selects acupoints based on the functional characteristics of each stage. In the oral stage, acupoints such as Lianquan and Jinjin are selected to improve tongue muscle movement, achieving an effectiveness rate of 93.3%; in the swallowing stage, acupoints such as Fengchi and Renying are selected to enhance the gag reflex, achieving an effectiveness rate of 90.9%; and in the esophageal stage, acupoints such as auricular acupuncture on the esophagus and Zusanli are selected to regulate sphincter function, achieving an effectiveness rate of 88.9%. This design of matching acupoints according to the stage significantly improves the precision and effectiveness of acupuncture treatment.

[0050] refer to Figure 4 This chart clarifies the importance and priority of each indicator in the multimodal swallowing assessment system. The UES open area has the highest weight (30%) because it directly reflects the coordinated function of the upper esophageal sphincter and is a core indicator for assessing swallowing safety; the hyoid bone displacement amplitude has a weight of 25%, which quantifies laryngeal elevation ability and is associated with pharyngeal food clearance efficiency; the FA value has a weight of 20%, which assesses the integrity of the dorsolateral medullary white matter pathway and reveals the neuropathological basis of swallowing disorders; the cough peak flow rate has a weight of 15%, used to measure the patient's compensatory clearance ability after aspiration; and the PAS score has a weight of 10%, which helps to determine the severity of aspiration. This weighting allocation is based on the contribution of each indicator to the swallowing function assessment, ensuring that key information is focused when integrating multimodal data, thus improving the accuracy and effectiveness of the assessment.

[0051] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multimodal swallowing function intelligent evaluation and intervention integrated system, characterized in that, Comprise the following modules: Multimodal data acquisition module, swallowing function image, physiological signal, clinical scale data; image data through television fluoroscopy, fiber endoscopy swallowing function examination, 3.0T MRI equipment acquisition; physiological signal data through surface electromyography, ultrasonic equipment acquisition; clinical scale data through system interaction interface input, all data correlation patient unique identification, storage format is DICOM, EDF, JSON; Multimodal intelligent evaluation module, containing image, physiological signal, scale score, machine learning fusion four units; image analysis unit calculates the amplitude of hyoid displacement, UES opening peak area, corticobulbar tract FA value; physiological signal analysis unit calculates the cortical-laryngeal muscle transmission latency and laryngeal elevation time; scale score unit grades DOSS-S scale; machine learning fusion unit adopts XGBoost algorithm, outputs swallowing disorder degree classification, aspiration risk probability and nutritional risk level, and model training uses 5-fold cross validation; Acupuncture treatment evaluation and adaptation module, including four units of staging evaluation, acupoint selection, operation parameter configuration and effect evaluation; the staging evaluation unit divides the swallowing disorder into oral cavity stage, swallowing stage and esophagus stage and grades them correspondingly; the acupoint selection unit selects corresponding acupoints according to the stages; the operation parameter configuration unit sets the needle insertion depth, needle retention time and special operation of acupoints in each stage; the effect evaluation unit evaluates the effect through Katakura drinking water test and standard scale; Combined rehabilitation intervention module, including three units of ice acid stimulation, standard swallowing training and neuromuscular electrical stimulation; the ice acid stimulation unit configures lemon acid concentration gradient, sets stimulation temperature, site, time length and frequency; The standard swallowing training unit includes oral cavity, pharyngeal stimulation and eating adjustment training; the neuromuscular electrical stimulation unit adopts NMES technology, sets electrode position, frequency, pulse width, current intensity and training time length and frequency; Result feedback and system iteration module, generates multimodal evaluation and intervention report, records evaluation data after each intervention, adopts AES256 encryption and stores in distributed database; through new data iteration machine learning fusion model, retrain every quarter, when the accuracy decreases by more than 5%, trigger model emergency iteration. 2.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, Also included is a swallowing function composite score calculation module that calculates a swallowing function composite score S through the formula calculating a swallowing function composite score S for a patient, where S is the swallowing function composite score, is a weight for image assessment, is a weight for physiologic signal assessment, is a weight for scale score, and determined by a machine learning model trained on 3000 historical patient data; S1 is the image assessment score, S2 is the physiologic signal assessment score, and S3 is the scale score score. 3.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, Also include multimodal data quality verification module, which verifies the quality of the collected image data and physiological signal data; image data verification indicators include television fluoroscopy image definition, fiber endoscopy swallowing function examination mucosa visibility and MRI DTI image signal-to-noise ratio; television fluoroscopy image definition requires that the contrast agent boundary definition within the frame is greater than or equal to 90%; if it is lower than this value, it is prompted to reacquire and adjust the equipment focal length; The fiber endoscope swallowing function test mucosa visibility requirement is that the mucosa texture recognition degree is greater than or equal to 85%, and when the value is less than the requirement, the clean endoscope lens is prompted; the MRI DTI image signal-to-noise ratio requirement is greater than or equal to 20 dB, and when the value is less than the requirement, the scanning time is prolonged; the physiological signal data verification index includes the surface electromyogram signal-to-noise ratio and the ultrasonic image resolution, the surface electromyogram signal-to-noise ratio requirement is greater than or equal to 30 dB, and when the value is less than the requirement, the electrode position and the skin impedance are adjusted; the ultrasonic image resolution requirement is that the hyoid cartilage structure resolution degree is greater than or equal to 95%, and when the value is less than the requirement, the probe position and the frequency are adjusted, the data passing the verification enters the evaluation module, and the data not passing the verification is marked with a reason and guided to be reacquired. 4.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, The acupuncture treatment evaluation and adaptation module further includes an acupoint positioning navigation unit, which uses ultrasonic guidance to position the acupuncture acupoint, when the oral cavity stage acupoint is positioned, the ultrasonic probe is placed in the middle of the neck front, the soft tissue layer under the hyoid bone corresponding to the Quanliao acupoint is displayed, and the needle insertion point and depth are marked; when the Fengchi acupoint is positioned in the swallowing stage, the ultrasonic probe is placed under the occipital bone at the back of the neck, the soft tissue outside the foramen magnum of the occipital bone corresponding to the Fengchi acupoint is displayed, and the needle insertion direction and angle are determined; when the ear needle esophageal acupoint is positioned in the esophagus stage, a high-frequency ultrasonic probe is placed in the ear cavity area, the cartilage and soft tissue junction corresponding to the esophageal acupoint is displayed, and the point stimulation position is marked; meanwhile, the ultrasonic image in the needle insertion process is displayed in real time. 5.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, Also included is a rehabilitation effect prediction module that predicts the effective probability of the rehabilitation intervention through a formula P = a * (Spost - Spre) + b * t + c P = a * (Spost - Spre) + b * t + c where P is the effective probability of the intervention, t is the duration of the intervention, Spre is the pre-intervention comprehensive score of swallowing function, Spost is the post-intervention comprehensive score of swallowing function, a is a time coefficient, b is a score change coefficient, and c is a constant term. When P≥0.8, it is determined that the intervention program is suitable, and the current program is continued to be executed. When 0.5≤P<0.8, the intervention parameters are adjusted. When P<0.5, the intervention program is re-evaluated and replaced. 6.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, The joint rehabilitation intervention module further includes an aspiration risk dynamic monitoring unit, which acquires physiological signals of a patient during swallowing in real time, including a swallowing reflex latency of a surface electromyogram, a laryngeal elevation delay time monitored by ultrasonic monitoring, and a blood oxygen saturation; when the swallowing reflex latency is greater than 200 ms, the laryngeal elevation delay time is greater than 400 ms, and the blood oxygen saturation decreases to below 94%, real-time early warning is triggered, and medical staff is informed through a system interface and a sound alarm. 7.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, Further including a nutritional support adaptation module, which configures a nutritional scheme according to an NRS2002 nutritional risk score and a serum prealbumin level, when the NRS2002 score is greater than or equal to 3 points or the serum prealbumin is less than 180 mg / L, it is determined that there is high nutritional risk, and the daily nutritional intake is calculated as body weight x 35 kcal; when the NRS2002 score is 1-2 points and the serum prealbumin is 180-220 mg / L, it is determined that there is moderate nutritional risk, and the daily nutritional intake is calculated as body weight x 30 kcal; when the NRS2002 score is 0 point and the serum prealbumin is greater than or equal to 220 mg / L, it is determined that there is low nutritional risk, and the daily nutritional intake is calculated as body weight x 25 kcal; meanwhile, the food consistency is adjusted according to the degree of dysphagia, IDDSI level 4 thick liquid is used for DOSS-S1-2 levels, IDDSI level 3 paste food is used for 3-4 levels, and IDDSI level 2 slightly thickened liquid is used for 5-7 levels. 8.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, The acupuncture treatment evaluation and adaptation module further comprises a stimulation intensity dynamic adjustment unit, which adjusts the acupuncture stimulation intensity by the formula I = I0 + k * (t - t0) I0 is the initial stimulation intensity; k is the adjustment coefficient; t is the acupuncture treatment time; I0 is the initial stimulation intensity; k is the adjustment coefficient; t is the acupuncture treatment time; the difference between the comprehensive swallowing function score at the moment and the score before treatment, the integral term represents the cumulative amount of score change during treatment; when I decreases when the cumulative amount increases; when I remains or slightly increases when the cumulative amount is unchanged or negative. 9.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, It also includes a cross-center data synchronization module that supports data synchronization of multi-center medical institutions, and the synchronization content includes patient multi-modal assessment data, intervention programs, and effect records. A distributed data synchronization protocol is used, each center sets up a data node, and a synchronization request is triggered within 5 seconds after data changes. SHA-256 algorithm is used to calculate data hash value before synchronization, and only the difference data with inconsistent hash value is synchronized. Offline synchronization is supported, data is stored in a local encrypted directory when offline, and is associated with encryption using device lock screen password. After networking, synchronization is automatically initiated. Data synchronization permissions are set, and center administrators configure synchronization ranges. Synchronization logs are recorded, including synchronization time, data volume, and synchronization status. 10.The multi-modal swallowing function intelligent evaluation and intervention integrated system according to claim 1, characterized in that, It also includes a user management and permission control module that divides users into three roles: patients, medical staff, and system administrators. The patient role only views personal assessment reports, intervention programs, and effect records, and submits feedback. The medical staff role collects data, performs assessments, configures intervention programs, and views patient data under their jurisdiction. Medical staff must log in through professional qualification certification and facial recognition. The system administrator role configures system parameters, manages user permissions, maintains databases, and performs model iteration. Operation log records are set, recording all user actions, and logs are saved for at least 3 years. Role-based access control strategy is used to prohibit unauthorized operations.

Images