Spinal cord injury respiratory function training method and system based on ultrasonic intention recognition

Abstract

The invention relates to the technical field of medical training systems, in particular to a spinal cord injury respiratory function training method and system based on ultrasonic intention recognition. A machine learning model is used for recognizing the breathing intention of trying to inhale and trying to forcibly exhale and cough of a patient, relevant information is converted into audio and video biological feedback signals, layered training such as basic activation endurance coordination and functional tasks is provided for the patient, muscle group signals in the training process are collected in real time for evaluation, and parameters are dynamically and finely adjusted; the system comprises a physiological signal acquisition module, an intention recognition processing module, a biological feedback output module, an interaction module, an equipment control execution module and a closed-loop training regulation module, and realizes intention-driven physiological respiration assistance. According to the invention, deep respiratory muscle accurate monitoring personalized training and man-machine cooperative assistance can be realized, the rehabilitation effect is improved, and the patient is assisted to recover the autonomous respiration ability.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of medical training system, and particularly relates to a spinal cord injury respiratory function training method and system based on ultrasound intention recognition. BACKGROUND

[0002] After high spinal cord injury, patients often have paralysis or severe functional decline of core respiratory muscles such as diaphragm, intercostal muscle and abdominal muscle, causing respiratory dysfunction, resulting in patients needing to rely on a ventilator for a long time to maintain life, and restoring self-breathing function is the core goal of rehabilitation treatment for such patients. In the field of respiratory function rehabilitation, the traditional training method has obvious limitations: on the one hand, the deep respiratory muscle group is located deep, and the traditional monitoring technology is difficult to accurately capture its activity state, and it is even more impossible to identify the respiratory intention corresponding to the early weak muscle contraction of the patient; on the other hand, the existing training lacks intuitive and real-time feedback mechanism, and the patient cannot perceive the activity effect of his own respiratory muscle, resulting in insufficient training targeting and weak motivation, and the effect of neural reeducation is limited. The ultrasound intention recognition technology has the core advantages of non-invasiveness, real-time and visualization, can accurately target the deep core respiratory muscle group, clearly capture its dynamic changes, and can also identify the weak muscle structure changes that cannot be detected by naked eye or palpation, and then accurately decode the respiratory intention, providing key technical support for respiratory training effect evaluation and active training.

[0003] However, there is no systematic respiratory function training scheme based on ultrasound intention recognition technology at present, and the existing rehabilitation means cannot fully exert the advantages of the technology in deep muscle group monitoring, intention decoding and biological feedback. Therefore, it is urgent to develop a spinal cord injury respiratory function training method and system based on ultrasound intention recognition, accurately evaluate the dynamic activity of respiratory muscle in a visualized manner, build a scientific biological feedback training mode, improve the training motivation of patients and the effect of neural reeducation, and help patients recover the ability of autonomous respiration.

[0004] Therefore, the present application is proposed. SUMMARY

[0005] In order to solve the above technical problems existing in the prior art, the present application provides a spinal cord injury respiratory function training method and system based on ultrasound intention recognition.

[0006] To achieve the above purpose, the technical scheme of the present application is as follows: In a first aspect, the spinal cord injury respiratory function training method based on ultrasound intention recognition comprises: S1, targeting the deep core muscle group based on an ultrasound detection device, collecting ultrasound images and signal characteristics of the core muscle group; S2, associating and matching the ultrasound images and signal characteristics with the respiratory related intention instructions of the patient through a machine learning model; S3, input the respiratory muscle activity information captured by the ultrasonic detection device into the biofeedback generator, and convert it into intuitive feedback signals through a multimodal output interface; S4, the patient performs targeted respiratory function training and visualized auxiliary respiratory regulation according to the intuitive feedback signals; S5, the ultrasonic detection device collects muscle group signals in the training process of the patient in real time to form real-time feedback information, guides the patient to perform active respiratory training, and realizes precise evaluation of the training effect.

[0007] Further, in step S1, the ultrasonic detection device includes an ultrasonic probe array, and the deep core muscle group includes diaphragm and abdominal muscle. The ultrasonic image and signal features include displacement curve slope, thickness change rate, and specific texture pattern signal of the muscle group.

[0008] Further, in step S2, the breathing-related intention instruction includes “try to inhale” and “try to force exhale / cough”. Further, in step S2, the breathing-related intention instruction includes “try to inhale” and “try to force exhale / cough”.

[0009] Further, in step S3, the intuitive feedback signal includes visual signal and auditory signal, which are output cooperatively; the respiratory muscle activity information includes intention-driven activity information; the patient receives the visual signal and auditory signal through a patient interaction terminal interface, and the patient interaction terminal interface includes a game area and a physiological parameter area. The game area includes a hot air balloon height interactive element, an energy bar interactive element, and a fire flame blowing out interactive element, and the physiological parameter area includes diaphragm thickness degree percentage, displacement distance degree percentage, abdominal muscle activation degree percentage, and coordination index percentage.

[0010] Further, the hot air balloon height interactive element corresponds to the diaphragm contraction strength in real time through the hot air balloon height, and is accompanied by an auditory signal: the higher the diaphragm contraction strength, the higher the hot air balloon rises, and the higher the output tone, and a clear confirmation sound is emitted when the contraction meets the standard; the energy bar interactive element is used to display the endurance points of continuous inhalation time, and is accompanied by an auditory signal: a stable sound effect is output during the accumulation of points, and a progressive prompt sound is emitted when the points reach the stage target. The fire flame blowing out interactive element is used to determine the flame extinguishing speed through the abdominal muscle contraction strength when the patient exhales, and is accompanied by an auditory signal: the greater the abdominal muscle contraction strength, the faster the flame extinguishing speed, and the more obvious the sound effect intensity, and a completion sound effect is emitted when the flame is completely extinguished.

[0011] Further, in step S3, the biofeedback generator synchronously converts and transmits the respiratory muscle activity information to the therapist monitoring interface through the multi-modal output interface, and the therapist monitoring interface includes a dual-view ultrasound module, a dynamic waveform module, an alarm system module and an AI suggestion module.

[0012] Further, the dual-view ultrasound module is used to display the diaphragm B-ultrasound information and the abdominal muscle elastography information side by side, the dynamic waveform module is used to draw a muscle activity timing curve, the alarm system module is used to automatically detect respiratory abnormal events, and the AI suggestion module is used to automatically generate training program optimization suggestions.

[0013] Further, the targeted respiratory function training in step S4 includes at least one of the following training modes: Basic activation training: for patients with complete paralysis or weak contraction, under the coordinated prompts of visual signals and auditory signals, the patient is guided to recognize and generate the contraction intention and structural changes of the target muscles, including the diaphragm and specific abdominal muscles; Endurance training: set gradually increasing muscle group displacement, thickness or thickening rate targets, require the patient to maintain the target for a specific duration, and output positive sound effects continuously when the maintenance is up to standard, while real-time feedback of the maintenance state is output through auditory signals; Coordination training: train the patient to actively lower the diaphragm during inhalation and sequentially activate the deep abdominal muscles during exhalation to avoid overcompensation of the auxiliary respiratory muscles, and display the muscle activation sequence through visual signals and prompt whether the activation timing is synchronized through auditory signals; Functional task training: simulate the coughing action, and real-time feedback whether the abdominal muscle contraction intensity and timing reach the effective coughing threshold, and output reinforcement sound effects when the intensity meets the standard, and output reminder sound effects when the timing is abnormal.

[0014] Further, in the endurance training, the target maintenance is achieved by allowing the patient to maintain the balloon in the interface at a certain height for a certain time, and the auditory signal is stably output during the balloon is maintained at the target height, and the sound effect is weakened or a warning sound is emitted when the target height is deviated.

[0015] Further, the training method can be combined with a diaphragm pacemaker to assist the patient in inhaling through a diaphragm pacemaker triggered electric stimulation pulse; and the training system corresponding to the training method can be combined with a smart respirator to realize breathing assistance pressure support under intention recognition.

[0016] In a second aspect, a spinal cord injury respiratory function training system based on ultrasound intention recognition includes: A physiological signal acquisition module is used to target deep core muscle groups to acquire ultrasound images and signal characteristics, and real-time acquisition of muscle group signals during patient training provides basic data support for intention recognition, feedback output and device control; An intention recognition processing module is configured to receive the ultrasound image and signal features output by the physiological signal acquisition module, associate and match the ultrasound image and signal features with the respiratory-related intention instruction of the patient through a machine learning model, and output an intention recognition result. A biofeedback output module is configured to receive the respiratory muscle activity information captured by the physiological signal acquisition module and the intention recognition result output by the intention recognition processing module, convert the respiratory muscle activity information and the intention recognition result into intuitive feedback signals through a multi-modal output interface, and output the intuitive feedback signals. An interaction module includes a patient interaction terminal and a therapist monitoring terminal. The patient interaction terminal is configured to present the intuitive feedback signals output by the biofeedback output module to the patient for training and regulation. The therapist monitoring terminal is configured to synchronously display training-related information and a device working state, thereby providing monitoring support for a therapist. A device control execution module is configured to receive the intention recognition result output by the intention recognition processing module in response to an assisted breathing demand, and control an external assisted breathing device to perform a corresponding assisted action, such as controlling a diaphragm pacemaker to trigger an electric stimulation pulse to assist inhalation or controlling an intelligent ventilator to perform pressure support for breathing assistance under the intention recognition. A closed-loop training regulation module is configured to receive the training process muscle group signals collected in real time by the physiological signal acquisition module and the working state signals of the device control execution module, and form real-time feedback information.

[0017] Compared with the prior art, the above-mentioned spinal cord injury respiratory function training method and system based on ultrasound intention recognition provided by the present application includes the following steps: target acquisition of ultrasound images and displacement curve slopes and other signal features of diaphragm and abdominal muscles through an ultrasound detection device, identification of respiratory intention of the patient trying to inhale and trying to forcefully exhale and cough by means of a CNN plus LSTM hybrid model, conversion of related information into audio and video biofeedback signals, provision of basic activation endurance coordination and functional task hierarchical training for the patient, real-time collection of muscle group signals during the training process for evaluation and dynamic fine-tuning of parameters. The training system includes physiological signal acquisition, intention recognition processing, biofeedback output, interaction, device control execution, and closed-loop training regulation modules, and can be linked with a diaphragm pacemaker and an intelligent ventilator to realize physiological respiratory assistance driven by intention. The present application realizes precise monitoring of deep respiratory muscles, individualized training, and human-machine collaborative assistance, thereby improving the rehabilitation effect and helping the patient to recover the ability of independent breathing. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A flowchart of the spinal cord injury respiratory function training method provided by the embodiment of the present application is shown in the figure. Figure 2 An architecture diagram of the spinal cord injury respiratory function training system provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0019] The technical solutions of the present application will be described clearly below in conjunction with the drawings. Obviously, the described embodiments are not all the embodiments of the present application, and all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present application.

[0020] It should be noted that, unless otherwise specified, the relative arrangement, numerical expression of the components and steps set forth in these embodiments should not be understood as limiting the scope of the present application.

[0021] The following description of the exemplary embodiments is merely illustrative in nature and is in no way intended to limit the application or its application or use in any way. Techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail herein, but should be considered part of the present specification when applicable.

[0022] Embodiment one Reference Figure 1 , Figure 1 The flowchart of the spinal cord injury respiratory function training method based on ultrasonic intention recognition proposed by the present application can be combined with a diaphragm pacemaker to assist the patient in inhaling by triggering an electric stimulation pulse through the diaphragm pacemaker. The training method corresponding to the training system can be combined with an intelligent respirator to realize respiratory assistance pressure support under intention recognition. The specific steps can include: S1, deep core muscle group ultrasonic image and signal feature acquisition: based on an ultrasonic detection device targeting the deep core muscle group, the ultrasonic image and signal features of the core muscle group are collected; the ultrasonic detection device includes an ultrasonic probe array, and the deep core muscle group includes the diaphragm and the abdominal muscle; the ultrasonic image and signal features include the displacement curve slope, thickness change rate, and specific texture pattern signal of the muscle group. Specifically, it includes: S11, complete device preparation and calibration before training. The ultrasonic detection device uses an ultrasonic probe array containing two high-frequency linear array probes (frequency 10-15 MHz), the core targeted muscle group is the diaphragm and the abdominal muscle, the device is ensured to establish real-time communication with the edge computing unit through wired transmission, the sensitivity of the ultrasonic probe is calibrated before collection to avoid signal distortion.

[0023] S12, the patient takes a semi-recumbent position (bed head raised 30°-45°), evenly applies ultrasonic coupling agent to the contact area between the ultrasonic probe and the skin to eliminate the interference of air gap on the signal, and then performs probe positioning: For diaphragm, one probe is fixed at the 8-10 intercostal line of the patient's anterior axillary line, and the subcostal or intercostal section M-mode ultrasound mode is used to clearly capture the dynamic changes of the diaphragm dome; for abdominal muscles, the other probe is fixed within a 5 cm range around the navel, focusing on the corresponding body surface area of the transversus abdominis and internal oblique muscle, and the B-mode ultrasound combined with elastography mode is used to ensure that the structural characteristics of the abdominal muscles during contraction can be accurately captured.

[0024] S13, start the ultrasonic detection device, set the acquisition parameters: frame rate 25 fps, image resolution 1024x768, signal sampling frequency 1 kHz, start collecting ultrasonic images and signal characteristics. Specifically, it includes: S131, ultrasonic image: diaphragm B-mode grayscale image (clearly showing muscle fiber distribution pattern), abdominal muscle elastography image (color-coded to show muscle strain level); S132, evaluation index (ultrasonic measurement), specifically including: S1321, displacement curve slope of muscle group: for diaphragm, measure the vertical movement distance of diaphragm dome during quiet breathing or forced inspiration (unit: centimeter) by M-mode ultrasound, calculate the displacement curve slope combined with time series data, the displacement of spinal cord injury patients after injury is often significantly reduced (less than 1 cm or abnormal movement occurs), and the slope value is also reduced; for abdominal muscles, measure the horizontal displacement distance of transversus abdominis in the respiratory cycle and calculate the corresponding time series slope; S1322, thickness change rate of muscle group: for diaphragm, measure the diaphragm thickness at the end of expiration and maximum inspiration, calculate the thickness change rate according to the formula, the specific formula is:

[0025] The value of healthy people is usually greater than 20-30%, and the TF of spinal cord injury patients after injury will be significantly reduced; for abdominal muscles, measure the thickness difference of transversus abdominis and other specific abdominal muscles during quiet expiration and forced expiration / cough, and calculate the corresponding thickness change rate; S1323, specific texture pattern signal: extract the muscle fiber texture distribution, gray level change characteristics and texture deformation pattern of diaphragm and abdominal muscles during contraction from ultrasonic images, form exclusive texture signals, and use them for feature matching in subsequent intention recognition. The raw data collected are transmitted to the edge computing unit in real time, after noise reduction by median filtering algorithm, the image data are standardized to uniform resolution grayscale images, and the signal data are converted to floating-point time series data, providing a high-quality basis for subsequent processing.

[0026] S2, breathing-related intention recognition and matching: through a machine learning model, the ultrasound image and signal features are associated and matched with the breathing-related intention instructions of the patient; the breathing-related intention instructions include "try to inhale" and "try to force exhale / cough"; it also includes the process of edge computing and extraction of data collected by the ultrasound probe array by the edge computing unit, the machine learning model includes an intention recognition AI engine for recognizing the ultrasound image and signal features of the patient and associating and matching the corresponding intention instructions. Specifically, it includes: S21, feature extraction and intention association preprocessing: the edge computing unit first separates the muscle area from the background and extracts a 256-dimensional texture feature vector (covering the muscle fiber distribution, gray scale change and contraction deformation pattern of the diaphragm / abdominal transverse muscle), a 64-dimensional signal feature vector (including diaphragm displacement curve slope, thickening rate, thickening speed, abdominal transverse muscle thickness change rate, contraction timing and coordination), all features come from the core indicators of real-time monitoring of ultrasound, ensuring direct correlation with breathing intention; then the features are standardized, physical quantities such as displacement and thickness are converted to values in the 0-1 interval to eliminate dimensional differences, and through a feature selection algorithm, core features strongly related to trying to inhale and trying to force exhale / cough (such as diaphragm thickening rate weight ratio 30%, abdominal transverse muscle contraction timing weight ratio 25%) are selected to reduce redundant information interference and lay the foundation for AI engine association and matching.

[0027] S22, calling a machine learning model (including an artificial intelligence engine) with a hybrid architecture of convolutional neural network + long short-term memory network (CNN+LSTM), achieving precise association with breathing intention through "pre-training mapping + real-time feature matching": First, based on a large number of labeled samples, a pre-training feature library (matching threshold 0.85) is constructed, the samples cover the feature combinations and corresponding intention labels of patients with different spinal cord injury degrees when trying to inhale and trying to force exhale / cough, and the model establishes exclusive association rules through learning; the association weight ratio of diaphragm-related features (displacement curve slope, thickening rate, thickening speed, texture pattern) to trying to inhale intention is ≥70%, and the association weight ratio of abdominal muscle-related features (thickness change rate, contraction timing, coordination, texture pattern) to trying to force exhale / cough intention is ≥75%; In real-time matching, the model first assigns weights to the feature data set extracted in S21, then compares the feature combinations in the pre-training feature library one by one, determines that trying to inhale requires diaphragm displacement curve slope >0.5 cm / s, thickening rate >10%, thickening speed ≥0.1 mm / s and texture contraction type, determines that trying to force exhale / cough requires abdominal transverse muscle thickness change rate >8%, contraction timing precedes abdominal rectus muscle, coordination meets the standard and texture meets the rapid contraction feature, and all conditions are met and the matching degree with the feature library is ≥0.85, the corresponding intention is confirmed.

[0028] S23, correlation result optimization and output: the AI engine first calculates the confidence (precision ± 0.01) of the intention correlation, if the confidence is ≥0.85, the corresponding intention type is output, if it is between 0.7-0.85, it is marked as a suspected intention and continues to monitor the next frame of feature data, if it is <0.7, it is determined as "no clear respiratory intention"; at the same time, combined with the integrity of the feature correlation and the core feature contribution value, the inspiratory effort intensity score (0-10 points, the more core features meet the requirements, the higher the contribution value, the higher the score) and the expiratory effort effectiveness index (0-100, abdominal muscle coordination correlation directly affects the index) are generated; In addition, the AI engine will record the correlation results and the actual training feedback of the patient in real time, adjust the feature weight through transfer learning (such as for a patient whose abdominal transverse muscle contraction timing feature correlation accuracy is low, the correlation weight of thickness change rate can be increased), and optimize the correlation rules combined with the individual calibration process (record the patient's baseline state and maximum effort state) to adapt to individual differences and improve long-term correlation accuracy.

[0029] S3, biofeedback signal generation and multi-end output: input the respiratory muscle activity information captured by the ultrasonic detection device into the biofeedback generator, and convert it into intuitive feedback signals through a multi-modal output interface; the intuitive feedback signals include visual signals and auditory signals, which are output cooperatively; the respiratory muscle activity information includes intention-driven activity information; the patient receives the visual signals and auditory signals through a patient interaction terminal interface, which includes a game area and a physiological parameter area. Specifically, it includes: S31, input the respiratory muscle activity information into the biofeedback generator, specifically including: Integrate the ultrasonic signal feature data after step S1 preprocessing, including the displacement curve slope of diaphragm and abdominal transverse muscle, thickness change rate and 256-dimensional texture feature vector, and then input the intention recognition result data output by step S2, including respiratory intention type, matching confidence, inspiratory effort intensity score and expiratory effort effectiveness index, while synchronously adding real-time state data such as current training mode identifier and training time, to form a complete comprehensive input data set; the edge computing unit encapsulates the comprehensive input data set into JSON format, and the time series data is split by 10ms per piece to ensure that data transmission and feedback output can be synchronized in real time; Data transmission preferentially adopts wired Ethernet, ensuring that the transmission rate is ≥1Mbps and the delay is ≤50ms, and Bluetooth 5.0 BLE is used as a backup transmission method, which is connected to the biofeedback generator through a standardized SPI bus interface (clock frequency 1MHz, data bits 8bits), and the generator identifies valid data frames through frame header (0xAA) and frame tail (0x55); The condition for triggering the input of data is the completion of each frame of ultrasonic data acquisition (25 fps, 40 ms / time) or the update of the intention recognition result (confidence > 0.85). After receiving the data, the generator automatically checks the integrity and rationality of the data. If there are missing fields, it will feedback the data retransmission instruction. If an abnormal value is detected (such as diaphragm thickening rate > 50%), it will trigger the median filter secondary correction to ensure that the input data is reliable and available.

[0030] S32, game area interaction element and feedback logic, the game area includes hot air balloon height interaction element, energy bar interaction element and fire blowing out interaction element; The hot air balloon height interaction element corresponds to the diaphragm contraction strength in real time through the hot air balloon height, and is matched with an audible signal: the higher the diaphragm contraction strength, the higher the hot air balloon rises, and the higher the output tone (range 200-800 Hz). When the contraction meets the standard (thickening rate ≥ 20%), a clear confirmation sound is emitted. The energy bar interaction element is used to display the endurance score of the continuous inhalation time, and is matched with an audible signal: a continuous stable sound effect is output during the accumulation of the score, and a progressive prompt sound is emitted when the score reaches the stage goal (2 seconds of continuous inhalation corresponds to 20 minutes).

[0031] S33, physiological parameter area display: the physiological parameter area includes diaphragm thickness percentage, displacement distance percentage, abdominal transverse muscle activation percentage and coordination index percentage, with the average level of healthy people as the 100% reference, and real-time refresh display.

[0032] S34, therapist monitoring end output: the biofeedback generator converts and transmits the respiratory muscle activity information to the therapist monitoring interface through the multi-modal output interface, and the therapist monitoring interface includes a double-view ultrasound module, a dynamic waveform module, an alarm system module and an artificial intelligence suggestion module; the double-view ultrasound module is used to display the diaphragm B-ultrasound information and the abdominal muscle elastography information side by side, the dynamic waveform module is used to draw the time sequence curve of muscle activity, the alarm system module is used to automatically detect respiratory abnormal events (such as blood oxygen < 90%, apnea > 10s), and the artificial intelligence suggestion module is used to automatically generate training scheme optimization suggestions (including fine-tuning direction) based on training data.

[0033] S4, respiratory function training and visualized regulation: the patient performs targeted respiratory function training and visualized auxiliary respiratory regulation according to the intuitive feedback signal; specifically, the targeted respiratory function training includes at least one of the following training modes: S41, basic activation training: for patients with complete paralysis or weak contraction, under the coordinated prompting of visual and auditory signals, guide the patient to recognize and produce the contraction intention and structural changes of the target muscle, including diaphragm and specific abdominal muscle; Specifically, the applicable object is complete paralysis or weak contraction (diaphragm thickening rate <10%, no obvious contraction of abdominal muscle), if the patient has a slight contraction (thickness change rate ≥3%), the simplified hot air balloon interface + 200Hz continuous prompt sound is started during operation, the patient tries to drive the diaphragm / abdominal transverse muscle to produce a slight contraction, and the interface value change is observed, 15 minutes each time, 2 times a day.

[0034] S42, endurance training: set gradually increasing muscle group displacement, thickness or thickening rate target, require the patient to maintain the target for a certain period of time, and output positive sound effect when the target is maintained; in endurance training, the target maintenance is realized by keeping the balloon in the interface at a certain height for a certain time, and a stable 500Hz medium tone (40dB) is output during the balloon is maintained at the target height; when the deviation is less than or equal to 10%, the tone is reduced to 400Hz; when the deviation is more than 10%, the tone is reduced or a warning sound is emitted.

[0035] Specifically, the initial target is diaphragm thickening rate ≥15%, maintaining 5 seconds, increasing the maintenance time by 2 seconds and the thickening rate target by 2% every week, the patient inhales to keep the hot air balloon at the target height, and the energy bar accumulates points synchronously, the sound effect is stable when the target is met, and the sound effect is reduced or a "drip-drip" warning sound is emitted when the target is deviated, 5 groups each time, 3 minutes interval between each group, 1 time a day.

[0036] S43, coordination training: training the patient to actively lower the diaphragm during inhalation and activate the deep abdominal muscle in sequence during exhalation to avoid overcompensation of the auxiliary respiratory muscle, and displaying the muscle activation sequence through visual signal and prompting whether the activation timing is synchronized through auditory signal; Specifically, the interface displays "abdominal transverse muscle→internal oblique muscle→rectus abdominis" color light bar (visual), matched with progressive enhancement sound effect (auditory), guiding the patient to actively lower the diaphragm (displacement ≥1cm) during inhalation and activate the deep abdominal muscle in sequence during exhalation, the therapist monitors the auxiliary respiratory muscle activity through the dynamic waveform module and issues a reminder when abnormal, 12 groups each time, 3 seconds of inhalation and 5 seconds of exhalation each group, 1 time a day.

[0037] S44, functional task training: simulate the action of coughing, real-time feedback whether the abdominal muscle contraction intensity and timing reach the effective cough threshold, output reinforced sound effect when the intensity meets the standard, and output reminder sound effect when the timing is abnormal.

[0038] Specifically, the flame blowout scene is started + "forceful exhalation" voice prompt, real-time feedback of abdominal muscle contraction strength and timing curve, strength ≥ 25% and timing consistent with "rapid contraction - last for 2 seconds - slow relaxation" to determine that the effective cough threshold is reached, strength meets to issue a reinforcement sound effect, timing anomaly issues a reminder sound effect, 10 groups per training, 2 minutes between each group, once a day.

[0039] S5, real-time acquisition of muscle group signals during patient training to form real-time feedback information to guide patients to actively perform respiratory training and achieve precise evaluation of training effect. Specifically, it includes: S51, real-time signal acquisition, the ultrasonic probe array continuously acquires diaphragm and abdominal muscle group signals at a frequency of 1 kHz.

[0040] S52, feedback information generation, training statistics update effective training time, number of times of meeting the standard, muscle function grade every 5 minutes, and the evaluation report containing completion degree, progress amplitude and items to be improved is generated through the patient terminal pop-up window display.

[0041] S53, training regulation and multi-dimensional fine-tuning: the evaluation report is synchronized to the therapist monitoring end, and the therapist adjusts the training target combined with artificial intelligence suggestions; multi-dimensional fine-tuning is realized based on the evaluation results, specifically including: S531, training target fine-tuning: if the diaphragm average thickening rate increases by ≥ 5% compared with the last training, the maintenance time of endurance training will be increased by 1-2 seconds in the next training; if the abdominal transverse muscle activation success rate increases by ≥ 10%, the intensity threshold of functional task training will be appropriately increased.

[0042] S532, auxiliary equipment parameter fine-tuning (if combined with auxiliary equipment): when combined with a diaphragm pacemaker, if the patient's independent diaphragm thickening rate is ≥ 25%, the electrical stimulation intensity will be reduced by 0.5-1 mA every 2 weeks, and the delay trigger time will be shortened from 0.2 seconds to 0.1 seconds, gradually increasing the patient's independent breathing dominance; when combined with an intelligent respirator, if the patient's effective inhalation effort (diaphragm displacement ≥ 1.5 cm) accounts for ≥ 70%, the pressure support level of the auxiliary control mode will be reduced by 1-2 cmH2O, or switched to the synchronized intermittent mandatory ventilation mode, and the offline transition will be optimized.

[0043] S533, biofeedback sensitivity fine-tuning: if the patient's training compliance is low (effective training time < 70% of the preset target), the interestingness of game-based feedback will be enhanced, such as increasing the hot air balloon rising speed and increasing the frequency of point rewards, while reducing the difficulty of the initial training target; if the patient's training effect is saturated (continuous 2 weeks of compliance rate ≥ 95%), the sensitivity of the feedback signal will be increased, such as narrowing the thickness change interval corresponding to the hot air balloon rising and increasing the cumulative difficulty of endurance points.

[0044] Embodiment Two Referring to Figure 2 , Figure 2 The framework of the spinal cord injury respiratory function training system based on ultrasonic intention recognition proposed by the present application comprises: M1, a physiological signal acquisition module, is used for targeting deep core muscle groups to acquire ultrasonic images and signal characteristics, and acquiring muscle group signals in real time during patient training to provide basic data support for intention recognition, feedback output and device control; Specifically, the ultrasonic probe array contains two high-frequency linear array probes (10-15 MHz), which target deep core muscle groups (diaphragm, abdominal muscle) to acquire ultrasonic images and signal characteristics. Muscle group signals are acquired in real time at a frequency of 1 kHz during training, and the data is transmitted to the edge computing unit and the intention recognition processing module through wired transmission to provide basic data support for subsequent processes.

[0045] M2, an intention recognition processing module, is used for receiving ultrasonic images and signal characteristics output by the physiological signal acquisition module, associating and matching the ultrasonic images and signal characteristics with patient's breathing-related intention instructions through a machine learning model, and outputting intention recognition results; Specifically, the machine learning model (containing an artificial intelligence engine) integrated in the embedded computing unit has a hybrid architecture of convolutional neural network + long short-term memory network (CNN+LSTM). It receives preprocessed data output by the physiological signal acquisition module, completes muscle region separation and feature vector extraction (256-dimensional texture features, 64-dimensional signal features), matches with the pre-trained feature library of "attempt to inhale" and "attempt to force exhale / cough" (initial threshold 0.85), and outputs intention type, matching confidence, and quantitative indicators such as "inhale effort intensity score" and "exhale effort effectiveness index".

[0046] M3, a biofeedback output module, receives respiratory muscle activity information captured by the physiological signal acquisition module and intention recognition results output by the intention recognition processing module, and converts them into intuitive feedback signals through a multi-modal output interface and outputs them; Specifically, a microcontroller is used to receive respiratory muscle activity information (including intention driving signals) from the physiological signal acquisition module and recognition results from the intention recognition processing module, and convert them into visual signals and auditory signals through HDMI video interface and 3.5mm audio interface (multi-modal output interface) for synchronous output to the interactive module.

[0047] M4, an interactive module, includes a patient interactive terminal and a therapist monitoring terminal. The patient interactive terminal is used to present intuitive feedback signals output by the biofeedback output module to the patient for training control. The therapist monitoring terminal is used to synchronize training-related information and device working status to provide monitoring support for therapists. Specifically, the patient interaction terminal is a 10.1-inch touch display screen, the interface is divided into a game area (including hot air balloon height, energy bar, and flame blowing out interactive elements) and a physiological parameter area (displaying diaphragm thickness percentage, displacement distance percentage, abdominal muscle activation percentage, and coordination index percentage), and a built-in loudspeaker realizes auditory signal output; the therapist monitoring terminal is a 19-inch industrial display screen, the interface includes a double-view ultrasound module, a dynamic waveform module, an alarm system module, and an artificial intelligence suggestion module, and training data and device working states are synchronized through Ethernet.

[0048] M5, a device control execution module, configured to receive an intention recognition result output by the intention recognition processing module in response to an assisted breathing demand, and control an external assisted breathing device to perform a corresponding assisting action: control a diaphragm pacemaker to trigger an electric stimulation pulse to assist inhalation, and control an intelligent respirator to perform a breathing assistance pressure support under the intention recognition; Specifically, a communication connection is established with the external assisted breathing device (diaphragm pacemaker and intelligent respirator) through a CAN bus protocol, the intention recognition result of the intention recognition processing module is received, a corresponding assisting action is triggered according to a preset logic, and intention-driven breathing assistance is realized.

[0049] M6, a closed-loop training regulation module, configured to receive a training process muscle group signal collected by the physiological signal collection module in real time and a working state signal of the device control execution module, and form real-time feedback information.

[0050] Specifically, the real-time muscle group signal of the physiological signal collection module and the working state signal of the device control execution module are integrated to generate training statistical data (effective training time, number of times of reaching the standard, etc.) and a quantitative evaluation report, and the intention recognition matching threshold and the assisted device control parameter are dynamically optimized to ensure training accuracy and closed-loop iteration.

[0051] The system workflow of the system specifically includes: After the system is started, each module is powered on to automatically complete communication connection and function self-checking, to ensure that the ultrasonic probe, the computing unit, the interactive terminal and other devices work normally. Subsequently, the physiological signal acquisition module completes target positioning of the ultrasonic probe, acquires ultrasonic images and signal characteristics according to preset parameters (frame rate 25 fps, resolution 1024*768), and transmits the acquired data to the intention recognition processing module in real time; the intention recognition processing module pre-processes and extracts features from the received data, matches the respiratory-related intention through a machine learning model, and outputs the recognition result and quantitative indicators; the biological feedback output module receives the respiratory muscle activity information and the intention recognition result, converts them into audio and video signals, presents them to the user through the patient interactive terminal, and synchronizes the related data to the therapist monitoring terminal; the device control execution module triggers the electrical stimulation pulse output of the diaphragm pacemaker or the mode switching and pressure adjustment of the intelligent respirator according to the intention recognition result; the closed-loop training regulation and control module continuously integrates real-time training data and device working status, generates feedback information and adjusts parameters accordingly, to realize dynamic adjustment and iteration of the whole process of the system. The specific adjustment logic is as follows: Adjustment of the physiological signal acquisition module: according to the clarity and stability of the real-time training signal, the acquisition parameters are adjusted, if motion artifacts appear in the ultrasonic image, the frame rate is increased to 30 fps to enhance the dynamic capture ability, if the signal noise is large, the signal sampling frequency is adjusted to 1.2 kHz and the noise reduction algorithm weight is strengthened to ensure the data acquisition quality; Adjustment of the intention recognition processing module: according to the accuracy and false trigger rate of intention recognition, the matching threshold is dynamically adjusted, if the false recognition rate of the "attempt to inhale" intention is higher than 5%, the matching threshold is increased from 0.85 to 0.88-0.90, if the recognition success rate is lower than 80%, the threshold is decreased to 0.80-0.83, and the feature extraction weight is optimized to enhance the priority of key features such as diaphragm thickening speed and abdominal muscle contraction timing; Adjustment of the device control execution module: combined with the patient training effect and the device working status, the auxiliary device parameters are adjusted, if the patient's independent contraction strength is improved (thickening rate >=25%) under the assistance of the diaphragm pacemaker, the electrical stimulation intensity is decreased by 0.5-1 mA every 2 weeks, if the patient's effective inhalation effort is increased (diaphragm displacement >=1.5 cm) under the assistance of the intelligent respirator, the pressure support level is gradually decreased by 1-2 cmH2O, to realize the smooth transition from "assistance to independent"; Adjustment after receiving the therapist regulation and control instruction: if the therapist issues a training target adjustment instruction (such as increasing the endurance training time, strengthening the abdominal transverse muscle activation training) through the monitoring terminal, the closed-loop training regulation and control module adjusts the corresponding parameters in real time, and synchronously updates the training target threshold of the patient interactive terminal and the sensitivity of the biological feedback signal (such as the mapping ratio of flame extinguishing speed and abdominal muscle contraction strength).

[0052] In summary, the present application has the following advantages: 1. Intention recognition layer, through machine learning model, associates ultrasound image / signal features with patient's "attempt to inhale" or "attempt to force exhale / cough" intention instructions, and outputs quantitative indicators to guide patient's autonomous respiratory training and provide reference for therapists; 2. Through real-time biofeedback training, ultrasound-captured respiratory muscle activity information (especially intention-driven activity) is converted into intuitive visual or auditory signal feedback to the patient in real time, which can guide the patient to consciously and accurately control the muscle training; 3. Using ultrasound to detect small muscle changes, not only can guide self-breathing training through visualization, but also can be used for "intention-driven" neural regulation and assistance through intention recognition. The recognized respiratory intention signal can control the auxiliary equipment (such as synchronous diaphragm pacemaker, intelligent respirator) in real time, and achieve more physiological respiratory support.

[0053] The above specific embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the examples, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for training respiratory function in spinal cord injury based on ultrasound intent recognition, characterized in that, include: S1. Based on ultrasound detection equipment, target deep core muscle groups and collect ultrasound images and signal characteristics of core muscle groups; S2. Using a machine learning model, the ultrasound images and signal features are associated and matched with the patient's breathing-related intention commands; S3. Input the respiratory muscle activity information captured by the ultrasound detection device into the biofeedback generator, and convert it into an intuitive feedback signal output through the multimodal output interface; S4. Based on intuitive feedback signals, patients undergo targeted respiratory function training and visual assisted respiratory control. S5. Ultrasound detection equipment collects muscle signals from patients during training in real time, generates real-time feedback information, guides patients in active breathing training, and enables accurate assessment of training effectiveness.

2. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 1, characterized in that, In step S1, the ultrasound detection device includes an ultrasound probe array, and the deep core muscle group includes the diaphragm and abdominal muscles. The ultrasound image and signal features include the slope of the muscle group displacement curve, the rate of change of thickness, and specific texture pattern signals.

3. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 1, characterized in that, In step S2, the breathing-related intention commands include "try to inhale" and "try to exhale forcefully / cough"; It also includes the process of edge computing extraction of data acquired by the ultrasound probe array through the edge computing unit. The machine learning model includes an intent recognition AI engine, which is used to identify the ultrasound image and signal features of the patient and associate them with the corresponding intent commands.

4. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 1, characterized in that, In step S3, the intuitive feedback signal includes visual and auditory signals, which are output in tandem; the respiratory muscle activity information includes intention-driven activity information; the patient receives the visual and auditory signals through a patient interaction terminal interface, which includes a game area and a physiological parameter area. The game area includes interactive elements for hot air balloon height, energy bar, and flame extinguishing; the physiological parameter area includes percentages for diaphragm thickness, displacement distance, transverse abdominis activation, and coordination index.

5. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 4, characterized in that, The hot air balloon height interactive element corresponds to the diaphragm contraction intensity in real time through the hot air balloon height, and is also combined with auditory signals: the higher the diaphragm contraction intensity, the higher the hot air balloon rises, the higher the output pitch, and a crisp confirmation sound is emitted when the contraction reaches the target. The energy bar interactive element is used to display the endurance score of continuous inhalation time, and is accompanied by auditory signals: a continuous and stable sound effect is output during the accumulation of the score, and a progressive prompt sound is emitted when the score reaches the stage target. The flame extinguishing interactive element is used to determine the flame extinguishing speed by the intensity of abdominal muscle contraction when the patient exhales, and is combined with auditory signals: the greater the intensity of abdominal muscle contraction, the faster the flame extinguishes, the more obvious the sound effect, and a completion sound effect is emitted when the flame is completely extinguished.

6. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 1, characterized in that, In step S3, the biofeedback generator synchronously converts and transmits respiratory muscle activity information to the therapist monitoring interface through a multimodal output interface. The therapist monitoring interface includes a dual-view ultrasound module, a dynamic waveform module, an alarm system module, and an AI suggestion module.

7. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 6, characterized in that, The dual-view ultrasound module is used to display diaphragmatic ultrasound information and abdominal muscle elastography information in parallel; the dynamic waveform module is used to draw muscle activity time-series curves; the alarm system module is used to automatically detect abnormal respiratory events; and the AI ​​suggestion module is used to automatically generate training program optimization suggestions.

8. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 1, characterized in that, The targeted respiratory function training in step S4 includes at least one of the following training modes: Basic activation training: For patients with complete paralysis or weak contraction, guided by the synergistic cues of visual and auditory signals, the patient is guided to identify and generate the intention to contract and structural changes of target muscles, including the diaphragm and specific abdominal muscles. Endurance training: Set progressively increasing muscle group displacement, thickness, or thickening rate targets, requiring patients to maintain the target for a specific duration, while providing real-time feedback on the maintenance status through auditory signals, and continuously outputting positive sound effects when the target is achieved; Coordination training: Train patients to actively lower the diaphragm during inhalation and activate the deep abdominal muscles in sequence during exhalation to avoid overcompensation by accessory respiratory muscles. Visual signals are used to demonstrate the order of muscle activation, and auditory signals are used to indicate whether the activation sequence is synchronized. Functional task training: Simulates coughing action, provides real-time feedback on whether the intensity and timing of abdominal muscle contraction have reached the effective cough threshold, outputs a reinforced sound effect when the intensity meets the standard, and outputs a reminder sound effect when the timing is abnormal.

9. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 8, characterized in that, In the endurance training, the goal is maintained by having the patient keep a balloon at a certain height for a certain period of time. While the balloon is kept at the target height, the auditory signal is output stably. When the balloon deviates from the target height, the sound effect weakens or a warning sound is emitted.

10. The method for training respiratory function in spinal cord injury based on ultrasound intent recognition according to claim 1, characterized in that, The training method can be combined with a diaphragm pacemaker to trigger electrical stimulation pulses to assist the patient's inhalation; the training system corresponding to the training method can be combined with a smart ventilator to achieve respiratory support pressure support under intention recognition.

11. A spinal cord injury respiratory function training system based on ultrasound intent recognition, characterized in that, include: The physiological signal acquisition module is used to acquire ultrasound images and signal features of deep core muscle groups, and to acquire muscle group signals in real time during patient training, providing basic data support for intent recognition, feedback output and equipment control. The intent recognition processing module is used to receive ultrasound images and signal features output by the physiological signal acquisition module, and to associate and match the ultrasound images and signal features with the patient's breathing-related intent commands through a machine learning model, and output the intent recognition result. The biofeedback output module receives respiratory muscle activity information captured by the physiological signal acquisition module and intent recognition results output by the intent recognition processing module, and converts them into intuitive feedback signals and outputs them through the multimodal output interface. The interaction module includes a patient interaction terminal and a therapist monitoring terminal. The patient interaction terminal is used to present the patient with intuitive feedback signals output by the biofeedback output module, which the patient can then train and adjust. The therapist monitoring terminal is used to synchronize training-related information and equipment operating status, providing monitoring support for therapists; The device control execution module is used to respond to assisted breathing needs, receive the intent recognition results output by the intent recognition processing module, and control the external assisted breathing device to perform corresponding assisted actions: control the diaphragm pacemaker to trigger electrical stimulation pulses to assist inhalation, and control the intelligent ventilator to provide assisted breathing pressure support under intent recognition. The closed-loop training control module is used to receive muscle group signals during the training process collected in real time by the physiological signal acquisition module and the working status signals of the equipment control execution module, forming real-time feedback information.

Images