Intelligent control method of tongue pressure feedback training instrument

Abstract

The invention provides an intelligent control method for a tongue pressure feedback training instrument, and the method comprises the steps: collecting tongue pressure peak data of a patient, a laryngeal electromyographic signal value of the tongue pressure feedback training instrument and a swallowing sound spectrum feature in real time, constructing a biological signal fusion data set containing time sequence information, and obtaining a biological signal time sequence distribution feature of the tongue pressure feedback training instrument; through the time window with the abnormally reduced pressure change rate, associating the frequency domain characteristic change of the swallowing sound spectrum obtained based on swallowing sound spectrum characteristic analysis, and obtaining the frequency band distribution information of sound abnormity; according to the frequency band distribution information of the abnormal sound, adjusting parameters needed by self-adaptive feedback of the tongue pressure feedback training instrument are calculated, and the adjusting amplitude of opening and closing of the air escape valve is obtained through a preset mapping relation; and inputting the deviation between the tongue pressure peak value and the target tongue pressure value and the adjustment amplitude of opening and closing of the air escape valve to a machine learning algorithm, and outputting a triggering optimization time point for guiding the patient to apply the target tongue pressure to a telescopic air bag of the tongue pressure feedback training instrument.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an intelligent control method for a tongue pressure feedback training device. Background Technology

[0002] In the field of medical rehabilitation, the development of intelligent training devices based on multimodal biosignals is of great significance, especially in the rehabilitation training of patients with swallowing dysfunction, where their role is irreplaceable. These devices capture multiple physiological signals from the human body, providing patients with precise feedback and guidance to help restore normal physiological function. However, current technologies still face many shortcomings in practical applications, particularly in their insufficient adaptability to individual patient differences in dynamic environments. Existing methods often struggle to capture and integrate the complex relationships between multiple signals in real time, limiting the accuracy and timeliness of feedback during training and affecting rehabilitation outcomes. A deeper analysis of this field reveals that the core challenge lies in effectively handling the temporal synergy between multiple physiological signals. In particular, the matching of changes in tongue pressure with the critical moments of swallowing has become a key area for technological breakthroughs. When the peak of tongue pressure fails to synchronize with the critical timing of swallowing, it leads to a series of chain reactions, such as abnormal rates of pressure change, which in turn affect the timing of laryngeal movements, ultimately interfering with the coordination of the entire swallowing process. This temporal inconsistency not only increases the difficulty of swallowing function rehabilitation training but also makes it difficult for devices to provide effective guidance. In traditional swallowing rehabilitation training, the mechanical design of the equipment has a decisive impact on the training effect. In particular, the precision and response speed of the pneumatic control system directly affect whether the patient can obtain accurate biofeedback. Furthermore, the deflation valve in the tongue depressor directly determines the rate of pressure change during training by adjusting the airbag pressure to simulate different resistance levels. The telescopic airbag in the tongue depressor directly contacts the patient's tongue, guiding the patient to apply target tongue pressure through changes in airbag pressure; its dynamic adjustment and real-time matching with tongue pressure changes are crucial to the training effect. Therefore, how to monitor the temporal relationship between tongue pressure changes and swallowing movements in real time during training, and dynamically adjust the device's feedback mechanism through intelligent algorithms, has become a critical issue that urgently needs to be addressed. Summary of the Invention

[0003] This invention provides an intelligent control method for a tongue pressure feedback training device, mainly comprising:

[0004] Real-time acquisition of patients' tongue pressure peak data, laryngeal electromyography signal values ​​of the tongue pressure feedback training device, and swallowing sound spectrum features; construction of a biosignal fusion dataset containing temporal information; and obtaining the biosignal temporal distribution characteristics of the tongue pressure feedback training device.

[0005] Based on the temporal distribution characteristics of the biosignal of the tongue pressure feedback training device, the temporal coordination deviation between the peak tongue pressure and the critical period of swallowing is analyzed by temporal correlation analysis to obtain the time difference between the two. If the time difference between the two exceeds the preset threshold range, it is determined that there is a synchronization abnormality.

[0006] For synchronization anomalies, the dynamic trend characteristics of the pressure change rate of the tongue are extracted. Combined with the real-time monitoring of the delay data during laryngeal elevation, the correlation between the dynamic trend characteristics and the delay data during laryngeal elevation is analyzed to obtain the time window of abnormal decrease in pressure change rate.

[0007] By using the time window of abnormal decrease in pressure change rate, and correlating the frequency domain feature changes of swallowing sound spectrum obtained from swallowing sound spectrum feature analysis, the frequency band distribution information of sound abnormality is obtained.

[0008] Based on the frequency distribution information of the abnormal sound, the adjustment parameters required for the adaptive feedback of the tongue pressure feedback training device are calculated, and the adjustment range of the opening and closing of the vent valve is obtained through the preset mapping relationship.

[0009] The deviation between the peak tongue pressure and the target tongue pressure value and the adjustment range of the deflation valve are input into the machine learning algorithm, and the output is the trigger optimization time point that guides the patient to apply the target tongue pressure to the telescopic air bladder of the tongue pressure feedback training device.

[0010] By triggering optimized timing points, the control commands for opening and closing the vent valve of the tongue pressure feedback trainer are dynamically adjusted, the output frequency of the guidance signal is updated synchronously, and the stability indicators during the swallowing training process are determined through real-time sensor data. This generates the final control strategy for opening and closing the vent valve of the tongue pressure feedback trainer and the timing pattern for triggering optimized timing points.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses an adaptive control method for a tongue depressor feedback training device. It utilizes multimodal sensors to collect real-time data on the patient's peak tongue pressure, laryngeal electromyography signals, and swallowing sound spectrum characteristics, constructing a biosignal fusion dataset containing temporal information. This invention analyzes the temporal coordination deviation between peak tongue pressure and the critical swallowing period, detects synchronization abnormalities, and, combined with the dynamic trend characteristics of tongue pressure change rate and laryngeal elevation delay data, determines the time window for abnormal pressure decrease. Furthermore, it correlates the frequency domain characteristics of the swallowing sound spectrum to obtain the frequency band distribution information of abnormal sounds. Based on the above analysis results, this invention calculates the adaptive feedback adjustment parameters of the tongue depressor feedback training device, optimizes the adjustment amplitude of the vent valve opening and closing, and uses a machine learning algorithm to determine the trigger optimization time point, dynamically adjusting the output frequency of the vent valve opening and closing control command and the guiding signal. This invention enables precise adaptive control of the tongue depressor feedback training device, improving the stability and effectiveness of swallowing training. Attached Figure Description

[0013] Figure 1 This is a flowchart of an intelligent control method for a tongue pressure feedback training device according to the present invention. Detailed Implementation

[0014] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0015] like Figure 1 The intelligent control method for a tongue pressure feedback training device in this embodiment may specifically include:

[0016] Step S101: Real-time acquisition of the patient's tongue pressure peak data, laryngeal electromyography signal values ​​of the tongue pressure feedback training device, and swallowing sound spectrum features; construction of a biosignal fusion dataset containing temporal information; and obtaining the biosignal temporal distribution characteristics of the tongue pressure feedback training device.

[0017] The patient's tongue pressure peak data, laryngeal electromyography (EMG) signal value, and swallowing sound signal are synchronously acquired using a multimodal sensor at a preset acquisition frequency. The tongue pressure peak data is timestamped, the EMG signal value is amplitude normalized, and the swallowing sound signal is subjected to short-time Fourier transform to obtain frequency domain spectral features. The three signal data are then time-aligned based on the timestamps to obtain time-aligned multimodal biosignal data. For the time-aligned multimodal biosignal data, a sliding window method is used to extract the tongue pressure peak change rate, EMG signal root mean square value, and swallowing sound dominant frequency component within each time window. If the tongue pressure peak change rate within a certain time window exceeds a preset threshold, that window is marked as a key time segment. The three signals within the key time segment are weighted and fused. The weight coefficient is calculated by dividing each signal by its variance within that time window to obtain the signal-to-noise ratio (SNR). The weight of each signal is determined based on its proportion to the sum of the SNRs of the three signals, resulting in the fused biosignal time sequence. Based on the fused biosignal time series, the K-means clustering algorithm is used to perform cluster analysis on the data points in the sequence. By calculating the sum of squares within each cluster under different numbers of clusters, the optimal number of clusters is determined at the inflection point where the rate of decrease of the sum of squares within each cluster slows down significantly as the number of clusters increases. The coordinates of the center point of each cluster and the distribution density of data points within each cluster are obtained. The Euclidean distance between the cluster center points between adjacent time windows is calculated as the offset distance, and the arctangent value of the difference in center point coordinates is calculated as the offset direction to determine the biosignal time series distribution characteristics of the tongue pressure feedback training device.

[0018] Specifically, the synchronous acquisition mechanism of multimodal sensors is the foundation for achieving accurate biosignal fusion.

[0019] Specifically, peak tongue pressure data was acquired using a pressure sensor placed between the patient's tongue and palate. The pressure applied by the tongue during swallowing generates electrical signal changes. Laryngeal electromyography (EMG) signals were obtained using surface EMG electrodes attached to the outer skin of the larynx, capturing the electrical activity of the laryngeal muscles during swallowing. Swallowing sound signals were acquired using a high-sensitivity microphone placed at an appropriate distance in front of the larynx. All three sensors operated at a uniform sampling frequency to ensure the data correspondence over time. Timestamps and signal preprocessing were crucial for subsequent analysis.

[0020] In one embodiment, each acquired data point is appended with a timestamp accurate to the millisecond level, providing a benchmark for subsequent timing alignment. EMG signal amplitude normalization, by calculating the maximum and minimum values ​​of the signal, maps the original signal to a standard range of 0 to 1, eliminating the influence of amplitude differences in EMG signals between different patients. Short-time Fourier transform converts the swallowing sound signal from the time domain to the frequency domain; the sound signal within each time window is decomposed into combinations of different frequency components, thereby extracting spectral information reflecting swallowing characteristics. The application of the sliding window method enables the capture of dynamic changes in biological signals.

[0021] It's important to note that the window size selection needs to balance temporal resolution and signal stability. A window that is too small may lead to signal instability, while a window that is too large may lose important transient features. The rate of change of peak tongue pressure reflects the velocity characteristics of tongue movement during swallowing; when the rate of change exceeds a preset threshold, it indicates that a critical stage of swallowing is underway. The root mean square value of the electromyographic signal represents the average intensity of muscle activity within this time window, while the dominant frequency component of the swallowing sound reflects the main sound frequency characteristics generated during swallowing. The weighting in the signal fusion process reflects the contribution of different signal qualities to the fusion result.

[0022] Preferably, the signal-to-noise ratio (SNR) is calculated as the ratio of variance to mean. Variance reflects the degree of signal fluctuation, while the mean represents the average level of the signal. The ratio of the two can effectively characterize the relative strength of useful information and noise in the signal. Weights are allocated according to the proportion of each signal's SNR to the total SNR, ensuring that higher-quality signals have a greater weight in the fusion result, thereby improving the reliability of the fused data. The application of the K-means clustering algorithm reveals the inherent distribution patterns of biological signals.

[0023] In one possible implementation, the sum of squares within clusters under different numbers of clusters is calculated, and its trend with increasing cluster number is observed. When the curve shows a clear inflection point, it indicates that further increasing the number of clusters no longer significantly improves data grouping; this number of clusters is the optimal value. Cluster center offset analysis further reveals the temporal evolution of biological signals. The Euclidean distance between center points of adjacent time windows reflects the magnitude of signal characteristic changes, while the offset direction indicates the trend of change. These temporal distribution characteristics provide a quantitative basis for assessing patients' swallowing function.

[0024] Step S102: Based on the biological signal temporal distribution characteristics of the tongue pressure feedback training device, the temporal coordination deviation between the peak tongue pressure and the critical period of swallowing is analyzed by temporal correlation analysis to obtain the time difference between the two. If the time difference between the two exceeds the preset threshold range, it is determined that there is a synchronization abnormality.

[0025] Based on the temporal distribution characteristics of biosignals, the time point of the peak tongue pressure is identified. By analyzing the joint change pattern of electromyographic signal amplitude and the dominant frequency component of swallowing sounds, when both exceed their respective preset thresholds, it is marked as the start time of the critical swallowing period. The peak tongue pressure time sequence and the critical swallowing period time sequence are extracted, and the time difference sequence between corresponding time points of the two sequences is calculated. For the time difference sequence, the temporal correlation between the peak tongue pressure time sequence and the critical swallowing period time sequence is calculated using the Pearson correlation coefficient. If the correlation coefficient is lower than a preset correlation threshold, the mean and standard deviation of the time difference sequence are obtained. The threshold range of normal temporal synergy is determined based on the mean plus or minus a preset multiple, and the judgment criterion for temporal synergy deviation is obtained. The judgment criterion is used to detect the real-time acquired biosignals, obtain the current peak tongue pressure time and the critical swallowing period time determined by the joint analysis of electromyographic signals and swallowing sounds, and calculate the time difference between the two. If the time difference exceeds the threshold range, it is determined that there is a synchronization abnormality.

[0026] Specifically, the temporal distribution characteristics of biological signals contain a wealth of swallowing function information.

[0027] Specifically, identifying the peak tongue pressure point relies on real-time monitoring of pressure sensor signals. When the pressure value reaches its local maximum, that moment is recorded as the peak point. These points reflect the instant when the patient exerts maximum force with their tongue during swallowing and are key indicators for assessing swallowing function. Determining the critical period of swallowing requires a comprehensive analysis of the coordinated changes in multiple biosignals.

[0028] In one embodiment, a sudden increase in the amplitude of the electromyographic signal indicates the onset of laryngeal muscle contraction, while a significant enhancement of the dominant frequency component of the swallowing sound indicates that food or liquid is passing through the pharynx. When both signals simultaneously exceed their respective preset thresholds, it signifies that the swallowing action has entered its most critical execution phase. This joint determination method is more reliable than single-signal determination and can effectively reduce the occurrence of misjudgments. The construction of the time difference sequence reveals the temporal relationship between tongue pressure and the critical period of swallowing.

[0029] It should be noted that during normal swallowing, the peak tongue pressure should occur slightly earlier than or simultaneously with the start of the critical swallowing period. This timing relationship ensures that food can be effectively propelled into the pharynx. By calculating the difference between the two moments in each swallow, the resulting sequence can reflect the overall state of the patient's swallowing coordination. The Pearson correlation coefficient is used here to assess the degree of synchronization between the two temporal sequences.

[0030] Preferably, the correlation coefficient is calculated based on a standardized time series, eliminating the influence of differences in swallowing frequency among different patients. When the correlation coefficient is close to 1, it indicates that the peak tongue pressure is highly synchronized with the critical period of swallowing; while a significantly decreased correlation coefficient suggests a coordination problem. The correlation threshold is usually set based on statistical data from healthy individuals, ensuring the objectivity of the judgment criteria. The threshold range is determined using statistical methods to establish normal intervals.

[0031] In one possible implementation, a large amount of normal swallowing time difference data is collected, and the mean is calculated to represent a typical temporal relationship, while the standard deviation reflects the degree of variability within the normal range. A preset multiple of the mean plus or minus the standard deviation is used as the upper and lower limits. This method considers individual differences while maintaining the stability of the judgment. The selection of the preset multiple needs to balance sensitivity and specificity; too small a multiple may lead to overdiagnosis, while too large a multiple may miss true abnormalities. The real-time detection process enables dynamic monitoring of swallowing synchronicity.

[0032] For example, if the peak tongue pressure during a swallow is detected to be significantly delayed beyond the critical period of swallowing, it indicates a lack of coordination between the tongue's pushing motion and the throat's contraction, potentially leading to food residue or aspiration. Conversely, if the peak tongue pressure occurs too early, it may mean that the tongue is exerting force prematurely before the throat is ready to receive food. The timely detection of such synchronization anomalies provides a clear direction for subsequent rehabilitation training and helps in developing targeted training programs.

[0033] Step S103: For synchronization anomalies, extract the dynamic trend characteristics of the pressure change rate of the tongue, combine them with the real-time monitoring of the delay data during larynx elevation, analyze the correlation between the dynamic trend characteristics and the delay data during larynx elevation, and obtain the time window of abnormal decrease in pressure change rate.

[0034] For detected synchronous abnormal signal segments, the ratio of pressure difference to time interval between adjacent sampling points is calculated from the tongue pressure sensor data to obtain a pressure change rate sequence. This sequence is smoothed using a moving average method, and the first-order difference value at each time point in the smoothed sequence is calculated as a dynamic trend feature. The laryngeal elevation initiation point is determined based on the moment of sudden amplitude increase in the laryngeal electromyography signal. The average time interval between laryngeal elevation and tongue pressure initiation during swallowing in healthy individuals is used as a standard time. The difference between the actual elevation time and the standard time is calculated to obtain laryngeal elevation delay data. A positive delay value indicates a lag in elevation. Using the dynamic trend feature and laryngeal elevation delay data, the correlation value is calculated using the Spearman correlation coefficient. When the dynamic trend feature is consistently negative and the correlation coefficient exceeds a preset threshold, this period is identified as an abnormal pressure drop segment. The start and end times of this abnormal drop segment are recorded to obtain the time window of the abnormal pressure change rate drop.

[0035] Specifically, the calculation of the rate of change of pressure is a fundamental indicator for assessing the coordination of tongue movements.

[0036] Specifically, by continuously collecting tongue pressure data points, the pressure difference between two adjacent moments is calculated, and then divided by the time interval to obtain the instantaneous rate of change. This calculation method can sensitively capture subtle changes in tongue force, especially the complete dynamic process of the tongue from rest to application of force and then to release during swallowing. The application of the moving average method eliminates interference from sensor noise and subtle muscle tremors by averaging multiple consecutive rate of change values, making the true pressure change trend clearer. The first-order difference value, as a dynamic trend feature, has a clear physical meaning.

[0037] In one embodiment, a positive first-order difference value indicates an increasing rate of pressure change, meaning the tongue is exerting force at a faster pace; a negative first-order difference value indicates a decreasing rate of pressure change, meaning the tongue is exerting force at a slower pace. This characteristic accurately reflects the acceleration and deceleration phases of tongue movement during swallowing, providing a quantitative basis for subsequent abnormality detection. Determining the timing of laryngeal elevation relies on the characteristic recognition of electromyographic signals.

[0038] It should be noted that the larynx undergoes a noticeable upward movement during swallowing, accompanied by the contraction of related muscle groups, which manifests as a sudden increase in amplitude on electromyography (EMG). By setting an appropriate threshold, the moment when the EMG amplitude exceeds that threshold can be determined as the onset of laryngeal elevation. This method is more objective and precise than visual observation. The process of establishing the standard time for healthy individuals demonstrates the application of statistical methods.

[0039] Preferably, by collecting swallowing data from a large number of healthy volunteers, the time interval between the initiation of tongue pressure and the elevation of the larynx during each swallow is recorded, and the average of these intervals is calculated as a standard reference. This standard moment represents the coordinated timing between tongue push and larynx reception during normal swallowing; any significant deviation from this standard may indicate an abnormality in swallowing function. The Spearman correlation coefficient is used here to assess the nonlinear relationship.

[0040] For example, if the rate of change in tongue pressure continues to decrease, and laryngeal elevation is also delayed accordingly, there is a positive correlation between the two. The strength of this correlation directly reflects the degree of coordination between tongue and laryngeal movements. A correlation coefficient close to 1 indicates high synchronization, while a coefficient close to 0 indicates relatively independent movements, potentially indicating problems with neural control or muscle coordination. Identifying the time window of abnormal pressure decrease has significant clinical implications.

[0041] In one possible implementation, a persistently negative dynamic trend indicates a continuous decrease in tongue pressure. If laryngeal elevation is delayed at this time and the correlation between the two exceeds a threshold, then an abnormality in swallowing coordination can be identified during this period. Recording the start and end times of this time window provides healthcare professionals with precise information on the timing of the abnormality, aiding in the analysis of its duration and severity. This allows for the development of targeted rehabilitation training plans, improving the patient's swallowing function recovery.

[0042] Step S104: By using the time window of abnormal decrease in pressure change rate, correlate the frequency domain feature changes of swallowing sound spectrum obtained based on swallowing sound spectrum feature analysis to obtain the frequency band distribution information of abnormal sound.

[0043] By analyzing the time window of an abnormal decrease in the rate of pressure change, swallowing sound signals collected within this time window are extracted. A Fast Fourier Transform (FFT) is performed on the sound signals to obtain spectral data. The spectrum is divided into multiple frequency bands according to preset frequency intervals, and the energy value within each band is calculated. Based on the energy value sequence, the average energy value of the same frequency band in a pre-established database of swallowing sound spectra from healthy individuals is compared as a standard energy distribution. The deviation of each frequency band's energy value from the standard value is calculated. If the deviation of a frequency band exceeds a preset deviation threshold, the frequency band is marked as an abnormal frequency band, and its frequency range and energy deviation value are recorded. Using the frequency range and energy deviation value of the abnormal frequency bands, data points are plotted on a two-dimensional coordinate system, where the horizontal axis represents the frequency value and the vertical axis represents the energy deviation degree. By connecting the data points of adjacent frequency bands to form a continuous curve, the regions exceeding the deviation threshold and their frequency ranges within the curve are identified, thus obtaining the frequency band distribution information of the abnormal sound.

[0044] Specifically, precise positioning of the time window is a key foundation for subsequent sound analysis.

[0045] Specifically, the time window of abnormally decreasing pressure change rate represents a specific period of weakened tongue thrust, often accompanied by swallowing coordination disorder. Extracting sound signals within this time range is significant because sound production is closely related to the movement of the tongue and larynx; when tongue pressure is abnormal, the corresponding sound characteristics also change. The application of Fast Fourier Transform in sound analysis converts the time-domain signal into a frequency-domain representation.

[0046] In one embodiment, the swallowing sound is represented in the time domain as an amplitude variation over time. After Fourier transform, the energy distribution of different frequency components can be clearly shown. Normal swallowing sounds have specific frequency characteristics, mainly concentrated in certain frequency bands, while this frequency distribution changes significantly during abnormal swallowing. The rationality of frequency band division directly affects the accuracy of abnormality detection.

[0047] It should be noted that the frequency range of human swallowing sounds is typically between tens of hertz and thousands of hertz. Dividing this range into fixed intervals, such as 100 hertz per band, provides sufficient frequency resolution. The energy value within each band is obtained by summing the squares of the amplitudes of all frequency components within that band, reflecting the intensity of the sound in that band. The process of establishing a spectral database of swallowing sounds from healthy individuals demonstrates the application of statistical methods.

[0048] Preferably, swallowing sounds from a large number of healthy volunteers under standard conditions are collected, and each sample undergoes the same spectral analysis to calculate the average and standard deviation of the energy in each frequency band. These statistical data form a benchmark for judging abnormalities; any significant deviation from this benchmark may indicate an abnormality in swallowing function. The calculation of the degree of energy deviation provides a quantitative indicator of abnormality.

[0049] For example, the difference between the measured energy value and the standard value of a certain frequency band, divided by the standard deviation, yields the standardized deviation. When this deviation exceeds a preset threshold, it indicates that the sound characteristics of that frequency band are significantly abnormal. This standardization process eliminates the influence of differences in energy reference values ​​across different frequency bands, making anomaly determinations for each band comparable. The construction of a two-dimensional coordinate graph visually displays the distribution characteristics of sound anomalies.

[0050] In one possible implementation, the horizontal axis represents continuous frequency values ​​from low to high frequencies, and the vertical axis represents the degree of energy deviation at each frequency point. By marking areas exceeding a threshold on the spectrum, it is possible to clearly identify which frequency bands are abnormal. The width of the region formed by consecutive abnormal frequency bands reflects the range of abnormal frequencies, while the peak value of the deviation indicates the frequency point with the most severe abnormality. This frequency band distribution information provides important information for clinical diagnosis. When abnormalities occur in the low-frequency band, it may be related to motor disorders at the base of the tongue; abnormalities in the mid-frequency band may reflect problems with pharyngeal contraction; and abnormalities in the high-frequency band are often related to turbulence caused by airflow passing through narrow channels. By analyzing the distribution patterns of abnormal frequency bands, medical staff can infer the specific type and severity of swallowing disorders, thereby developing more targeted rehabilitation plans.

[0051] Step S105: Based on the frequency distribution information of the abnormal sound, calculate the adjustment parameters required for the adaptive feedback of the tongue pressure feedback training device, and obtain the adjustment range of the opening and closing of the vent valve through a preset mapping relationship.

[0052] Based on the number and energy deviation of abnormal frequency bands in the frequency band distribution information of abnormal sounds, combined with the laryngeal elevation delay time value, a comprehensive abnormality index is calculated using a weighted summation method. This involves multiplying the number of abnormal frequency bands by a frequency band weighting coefficient, the energy deviation by an energy weighting coefficient, and the delay time value by a time weighting coefficient, and summing these three values ​​to obtain a comprehensive index value reflecting the overall degree of swallowing abnormality. Using this comprehensive index value, the system searches through stored standard adjustment parameter data, which includes vent valve control parameters corresponding to different comprehensive index values. A linear interpolation method is used to calculate the basic adjustment parameters corresponding to the current comprehensive index value. These basic adjustment parameters are then corrected based on the response speed in the patient's training history to obtain optimized adjustment parameter values. Using these optimized adjustment parameter values, the correspondence between them and the vent valve opening / closing angle is calculated. This correspondence is obtained by multiplying the optimized adjustment parameter value by an angle conversion coefficient to obtain the target opening / closing angle. The current opening / closing angle of the vent valve is then obtained, and the difference between the target angle and the current angle is calculated to obtain the adjustment range of the vent valve opening / closing.

[0053] Specifically, the calculation of the comprehensive anomaly index reflects the idea of ​​multi-dimensional assessment.

[0054] Specifically, the number of abnormal frequency bands directly reflects the extent to which swallowing sounds deviate from the normal pattern. When multiple frequency bands are abnormal simultaneously, it indicates that swallowing dysfunction involves multiple physiological processes. The degree of energy deviation quantifies the severity of each abnormal frequency band's deviation from the normal value; the greater the deviation, the more significant the change in the sound characteristics of that frequency band. The laryngeal elevation delay reflects problems with the coordination of neuromuscular control; the longer the delay, the more severe the temporal disorder between the initiation and execution of the swallowing reflex. The weighting coefficients are set based on the degree of influence of different indicators on swallowing function.

[0055] In one embodiment, the frequency band weighting coefficients are determined by considering the correlation between different frequency bands and the physiological mechanisms of swallowing. Low-frequency bands primarily reflect tongue root movement, mid-frequency bands correspond to pharyngeal contraction, and high-frequency bands are related to airflow through narrow passages. Corresponding weights are assigned based on the contribution of each frequency band to overall swallowing function. The energy weighting coefficient is determined based on the impact of energy deviation on swallowing safety; excessive energy deviation may increase the risk of aspiration. The time weighting coefficient reflects the importance of temporal coordination for swallowing efficiency. The establishment of standard adjustment parameter data involves the accumulation of a large amount of clinical data.

[0056] It should be noted that by collecting the optimal training parameters for patients under different comprehensive abnormality indices, a correspondence between the index values ​​and the vent valve control parameters was established. These parameters include the vent valve's opening speed, holding time, and closing speed, each of which directly affects the resistance change pattern during training. When the actual comprehensive index value falls between two known data points, the linear interpolation method calculates the appropriate parameter values ​​proportionally. The response speed in the patient's training history provides a basis for individualized adjustments.

[0057] Preferably, response speed is assessed by analyzing the patient's adaptation time to changes in resistance during previous training. Patients who adapt faster indicate better neuromuscular plasticity and can be trained with more aggressive parameters; patients who adapt slower require gentler parameter adjustments. This historical data-based correction makes the training program more tailored to individual characteristics. The physical significance of the angle conversion coefficient lies in establishing a quantitative relationship between the control signal and the mechanical action.

[0058] For example, the optimized value of the adjustment parameter represents the desired resistance adjustment intensity, while the opening and closing angle of the vent valve directly determines the size of the airflow channel, thus affecting the resistance during training. The angle conversion coefficient is determined through calibration experiments, ensuring an accurate mapping from control parameters to actual resistance. The calculation of the difference between the target angle and the current angle enables progressive adjustment.

[0059] In one possible implementation, a large difference indicates that the current training intensity is far from the patient's needs, requiring significant adjustments; a small difference necessitates only minor adjustments. This adjustment mechanism based on real-time feedback avoids abrupt changes in training intensity, ensuring the smoothness and safety of the training process. By precisely controlling the adjustment range of the vent valve, the most suitable training resistance for the patient's current swallowing function can be provided, thereby improving the effectiveness and efficiency of rehabilitation training.

[0060] Step S106: Input the deviation between the peak tongue pressure and the target tongue pressure value and the adjustment range of the vent valve opening and closing to the machine learning algorithm, and output the trigger optimization time point for guiding the patient to apply the target tongue pressure to the telescopic airbag of the tongue pressure feedback training device.

[0061] The difference between the real-time collected peak tongue pressure and the preset target tongue pressure value is calculated to obtain tongue pressure deviation data. This tongue pressure deviation data is combined with the adjustment range of the deflation valve to form a feature vector. Simultaneously, the current pressure value and pressure change rate of the telescopic balloon are obtained as supplementary features. The feature vector is input into a random forest algorithm. This algorithm is trained based on a pre-collected dataset of the time required for different patients to reach the target balloon pressure under various adjustment ranges. Through decision tree ensemble, the algorithm predicts the time value under the current feature conditions, obtaining the initial trigger time point. Based on the initial trigger time point and the current balloon pressure change rate, the ratio of the pressure change rate to the historical average change rate is calculated as a correction coefficient. The initial trigger time point is divided by the correction coefficient to obtain the adjusted time value. If the correction coefficient is greater than 1, the trigger time is shortened; if it is less than 1, the trigger time is extended, thus obtaining the optimized trigger time point for guiding the patient to apply the target tongue pressure to the telescopic balloon.

[0062] Specifically, the acquisition of tongue pressure deviation data reflects the gap between the patient's current ability and the training goals.

[0063] Specifically, the target tongue pressure value is the ideal training intensity determined based on the patient's swallowing function assessment results and rehabilitation stage, while the real-time collected peak tongue pressure represents the patient's current actual performance. The difference between the two is not just a simple numerical value, but also an important indicator reflecting the patient's training status. A positive deviation indicates that the patient is not applying enough pressure and needs more intensive training guidance; a negative deviation indicates that the patient is exerting too much force, which may lead to fatigue or discomfort. The construction of the feature vector embodies the idea of ​​multi-dimensional information fusion.

[0064] In one embodiment, the adjustment range of the deflation valve directly affects the rate of pressure release within the balloon; a larger adjustment range means faster pressure changes, requiring the patient to adjust tongue pressure more quickly. The current pressure value of the balloon provides information about the initial state of training, while the rate of pressure change reflects the dynamic response characteristics of the system. The combination of these features comprehensively describes the physical state during training. The advantage of the random forest algorithm in this application lies in its ability to handle complex nonlinear relationships.

[0065] It's important to note that the time it takes for different patients to reach the target pressure with the same adjustment amplitude varies. This variation is influenced by various factors, including the patient's muscle strength, neural reaction speed, and training proficiency. Random forests construct multiple decision trees, each trained based on different feature subsets and sample subsets. Finally, by integrating the predictions from multiple decision trees, a more accurate time estimate is obtained. The accumulation of historical datasets is fundamental to the algorithm's accuracy.

[0066] Preferably, during patient training, the system continuously records the characteristic parameters and the actual time required to reach the target pressure for each training session. This data, after cleaning and labeling, forms valuable training samples. As the amount of data increases, the algorithm can learn more patterns, including typical manifestations of different types of patients and temporal patterns under special circumstances. A pressure change rate correction mechanism ensures the accuracy of the triggering time.

[0067] For example, when the rate of change of airbag pressure is higher than the historical average, it indicates that the current system response is too fast, and the patient needs to begin applying pressure earlier to reach the target pressure at the appropriate time. The correction factor is calculated as a ratio; a value greater than 1 indicates that the system response is faster than average, thus requiring a shorter trigger time; a value less than 1 indicates that the system response is too slow, requiring a longer trigger time. This adaptive trigger time optimization results in a significant improvement in training effectiveness.

[0068] In one possible implementation, precise triggering time allows patients to apply pressure at the optimal moment, avoiding fatigue from applying pressure too early or missing the training window from applying pressure too late. Through real-time adjustments and optimization, personalized training guidance can be provided to each patient, improving training effectiveness and patient compliance, thereby accelerating the recovery process of swallowing function.

[0069] Step S107: By triggering the optimization time point, the control command for opening and closing the vent valve of the tongue pressure feedback trainer is dynamically adjusted, the output frequency of the guidance signal is updated synchronously, the stability index during the swallowing training process is determined by real-time sensor data, and the final control strategy for opening and closing the vent valve of the tongue pressure feedback trainer and the timing mode for triggering the optimization time point are generated.

[0070] By triggering optimized time points, commands for the opening / closing angle and duration of the vent valve at that moment are generated. These commands are converted into pulse width modulation signals and sent to the vent valve actuator. Simultaneously, the output frequency of the guidance signal is calculated based on the interval of the trigger time points, and the transmission frequency of the training prompt signal is adjusted to be consistent with the trigger frequency. Based on the tongue pressure value, airbag pressure value, and electromyography signal amplitude collected by real-time sensors, the ratio of the standard deviation to the mean of each signal within the time window is calculated as the coefficient of variation. The average of the three coefficients of variation is used as a stability index. If the stability index exceeds a preset stability threshold, the time period is recorded as an unstable training segment. The proportion of unstable training segments to the total training time is calculated to obtain the stability evaluation value of the training process. The stability evaluation value is used to optimize the adjustment parameters using gradient descent. The adjustment parameters include a gain coefficient that controls the response speed. When the stability evaluation value is lower than the preset stability target threshold, the gain coefficient is increased; when it is higher than the target threshold, the gain coefficient is decreased. After multiple training cycles, the control command sequence of the vent valve and the corresponding trigger time point sequence of each cycle are recorded. The control command pattern with the stability evaluation value closest to the target threshold is extracted to generate the final vent valve opening and closing control strategy and the timing pattern of the trigger optimization time point.

[0071] Specifically, the generation of vent valve control commands reflects the requirement for precise timing control.

[0072] Specifically, the opening angle command determines the size of the airflow channel, directly affecting the resistance intensity during training. When increasing training difficulty, the vent valve opening decreases, the pressure inside the cuff releases slowly, and the patient needs to apply greater tongue pressure to overcome the resistance; conversely, increasing the opening reduces the training difficulty. The duration command ensures the integrity of each training movement, avoiding insufficient training effect due to too short a control time, or patient fatigue due to too long a time. The application of pulse width modulation signals enables precise control of the vent valve.

[0073] In one embodiment, the average power of the vent valve actuator is controlled by adjusting the duty cycle of the pulse signal. A larger duty cycle results in higher average power for the actuator and faster valve operation. This control method offers better precision and response characteristics compared to simple on / off control, enabling continuous adjustment of the vent valve opening. Synchronous adjustment of the guide signal frequency ensures a consistent rhythm during patient training.

[0074] It should be noted that the training cue signals guide patients to apply tongue pressure at the correct time through sound or visual means. When the interval between trigger points is short, the frequency of the cue signal increases accordingly, guiding the patient to perform fast-paced training; when the interval is long, the frequency decreases, suitable for slow and deep training. This frequency synchronization mechanism helps patients establish a stable training rhythm. The coefficient of variation, as a core indicator for stability assessment, has clear statistical significance.

[0075] Preferably, the standard deviation reflects the degree of signal fluctuation within the time window, and the mean represents the average level of the signal. The ratio of the two eliminates the influence of different signal dimensions, allowing the stability of three different types of signals—tongue pressure, airbag pressure, and electromyography—to be assessed under the same standard. A smaller coefficient of variation indicates signal stability and a smooth training process; a larger coefficient of variation suggests the presence of irregular fluctuations. The calculation of the stability assessment value provides a quantitative indicator for the overall training quality.

[0076] For example, a high percentage of unstable training segments indicates that the patient has difficulty maintaining a stable movement pattern during training, possibly due to fatigue, lack of concentration, or unsuitable training parameters. This assessment not only reflects the quality of the current training but also provides a clear direction for subsequent parameter optimization. The application of gradient descent in parameter optimization ensures the convergence of the adjustment process.

[0077] In one possible implementation, the gain coefficient controls the system's response strength to deviations. When the stability assessment value is below the target threshold, it indicates that the system response is too sensitive; even small disturbances can cause large output changes. In this case, the gain coefficient needs to be increased to improve the system's damping characteristics. Conversely, when the assessment value is above the target, the system response is sluggish, and the gain coefficient needs to be decreased to improve sensitivity. The generation of the final control strategy and timing pattern is the result of multiple iterative optimizations. By recording the control command sequence and trigger time point sequence for each training cycle, the system can identify which control mode provides the best training stability for the patient. Selecting the mode with the stability assessment value closest to the target threshold as the final scheme ensures both the effectiveness of training and the stability of the process, providing patients with a personalized optimal training plan.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent control method for a tongue pressure feedback training device, characterized in that, The method comprises: Real-time acquisition of tongue pressure peak value data, laryngeal muscle electromyogram value of the tongue pressure feedback training instrument, and swallowing sound spectrum characteristics of the patient, construction of a biological signal fusion data set containing timing information, and obtaining biological signal timing distribution characteristics of the tongue pressure feedback training instrument; According to the biological signal timing distribution characteristics of the tongue pressure feedback training instrument, the timing correlation of the tongue pressure peak value and the timing cooperativity deviation of the swallowing key period is analyzed, the time difference between the two is obtained, and if the time difference between the two is detected to be out of the preset threshold range, it is judged that there is synchronization abnormality; For synchronization abnormality, the dynamic trend characteristics of the pressure change rate of the tongue are extracted, the correlation between the dynamic trend characteristics and the delay data of the laryngeal lifting is analyzed in combination with the real-time monitoring, and the time window of the abnormal decrease of the pressure change rate is obtained; Through the time window of the abnormal decrease of the pressure change rate, the frequency domain characteristic change of the swallowing sound spectrum obtained based on the swallowing sound spectrum characteristics is associated, and the frequency band distribution information of the sound abnormality is obtained; According to the frequency band distribution information of the sound abnormality, the adjustment parameters required for the adaptive feedback of the tongue pressure feedback training instrument are calculated, and the adjustment amplitude of the opening and closing of the air release valve is obtained through a preset mapping relationship; The deviation between the tongue pressure peak value and the target tongue pressure value and the adjustment amplitude of the opening and closing of the air release valve are input into the machine learning algorithm, and the trigger optimization time point of guiding the patient to apply the target tongue pressure to the stretchable air bag of the tongue pressure feedback training instrument is output. Through the trigger optimization time point, the control instruction of the opening and closing of the air release valve of the tongue pressure feedback training instrument is dynamically adjusted, the output frequency of the guide signal is synchronously updated, the stability index in the swallowing training process is determined through real-time sensor data, and the final air release valve opening and closing control strategy of the tongue pressure feedback training instrument and the timing mode of the trigger optimization time point are generated.

2. The intelligent control method of the tongue pressure feedback trainer according to claim 1, characterized in that, The real-time acquisition of the tongue pressure peak value data, the laryngeal muscle electromyogram value of the tongue pressure feedback training instrument, and the swallowing sound spectrum characteristics of the patient, the construction of a biological signal fusion data set containing timing information, and the obtaining of the biological signal timing distribution characteristics of the tongue pressure feedback training instrument comprise: The tongue pressure peak value data, the laryngeal muscle electromyogram value, and the swallowing sound signal are collected, a time stamp is added to the tongue pressure peak value data, the laryngeal muscle electromyogram value is subjected to amplitude normalization, and the swallowing sound signal is subjected to short-time Fourier transform to obtain frequency domain spectrum characteristics, and after timing alignment, multi-modal biological signal data are obtained; for the multi-modal biological signal data after timing alignment, a sliding window is used to extract the tongue pressure peak value change rate, the electromyogram root mean square value, and the swallowing sound main frequency component in each time window, the time window in which the tongue pressure peak value change rate exceeds the preset threshold is marked as a key timing segment, the signals in the key timing segment are subjected to weighted fusion, wherein the weight coefficients are determined based on the signal-to-noise ratios of the signals, and a fused biological signal timing sequence is obtained; according to the fused biological signal timing sequence, clustering analysis is used to determine the optimal cluster number, calculate the cluster center point coordinates and the offset distance, and obtain the biological signal timing distribution characteristics.

3. The intelligent control method of the tongue pressure feedback trainer according to claim 1, characterized in that, The biological signal time sequence distribution characteristics of the tongue pressure feedback training instrument are used to analyze the time sequence cooperation deviation of the tongue pressure peak value and the key period of swallowing through time sequence correlation, obtain the time difference between the two, and if the time difference is detected to be out of the preset threshold range, it is judged that there is a synchronization abnormality, including: According to the biological signal time sequence distribution characteristics, the tongue pressure peak time sequence is extracted, the joint change of the electromyographic signal amplitude and the main frequency component of the swallowing sound is used to mark the starting time of the key period of swallowing, and the time difference value sequence of the tongue pressure peak time sequence and the key period of swallowing time sequence is calculated. The time sequence correlation of the tongue pressure peak time sequence and the key period of swallowing time sequence is calculated through the correlation coefficient, and the threshold range is determined based on the mean and standard deviation of the time difference value sequence. For the real-time collected biological signal, the current tongue pressure peak time and the key period of swallowing time are obtained, the time difference between the two is calculated, the time difference is detected to be out of the threshold range, and it is judged that there is a synchronization abnormality.

4. The intelligent control method of the tongue pressure feedback trainer according to claim 1, characterized in that, For the synchronization abnormality, the dynamic trend characteristics of the pressure change rate of the tongue are extracted, including: From the signal segment of the synchronization abnormality, the ratio of the pressure difference value of the tongue pressure sensor data to the time interval is calculated to obtain the pressure change rate sequence. The pressure change rate sequence is smoothed, and the first-order difference value is calculated as the dynamic trend characteristics.

5. The intelligent control method of the tongue pressure feedback trainer according to claim 1, characterized in that, Through the time window of abnormal decrease of the pressure change rate, the frequency domain feature change of the swallowing sound spectrum obtained based on the swallowing sound spectrum feature analysis is associated to obtain the frequency band distribution information of the sound abnormality, including: The swallowing sound signal in the time window of abnormal decrease of the pressure change rate is extracted, and the frequency spectrum data is obtained by fast Fourier transform. The frequency band is divided, and the energy value of each frequency band is calculated. Compared with the average energy value of the swallowing sound spectrum of the healthy population, the frequency band whose deviation degree exceeds the threshold is marked as an abnormal frequency band. The frequency range and energy deviation value of the abnormal frequency band are recorded to obtain the frequency band distribution information.

6. The intelligent control method of a tongue pressure feedback trainer according to claim 1, wherein, According to the frequency band distribution information of the sound abnormality, the adjustment parameter required for the adaptive feedback of the tongue pressure feedback training instrument is calculated, and the adjustment amplitude of the opening and closing of the air release valve is obtained through the preset mapping relationship, including: According to the number and energy deviation degree of the abnormal frequency band in the frequency band distribution information of the sound abnormality, the comprehensive abnormality index is calculated combined with the throat lifting delay data. The adjustment parameter optimization value is generated according to the comprehensive abnormality index. The corresponding relationship between the adjustment parameter optimization value and the opening and closing angle of the air release valve is calculated through the adjustment parameter optimization value. The corresponding relationship is that the target opening and closing angle is obtained by multiplying the adjustment parameter optimization value by the angle conversion coefficient. The current opening and closing angle of the air release valve is obtained, the difference between the target angle and the current angle is calculated, and the adjustment amplitude of the opening and closing of the air release valve is obtained.

7. The intelligent control method of a tongue pressure feedback trainer according to claim 1, wherein, The deviation of the tongue pressure peak value and the target tongue pressure value and the adjustment amplitude of the opening and closing of the air release valve are input into the machine learning algorithm, and the trigger optimization time point of guiding the patient to apply the target tongue pressure to the stretchable air bag of the tongue pressure feedback training instrument is output, including: The difference between the tongue pressure peak value and the target tongue pressure value is calculated, the adjustment range of the opening and closing of the air release valve and the pressure value and the change rate of the stretchable air bag are combined to generate a feature vector; the feature vector is input into a random forest algorithm to predict a preliminary trigger time point; and the trigger optimization time point is obtained by adjusting the preliminary trigger time point and a correction coefficient of the pressure change rate.

8. The intelligent control method of the tongue pressure feedback trainer according to claim 1, characterized in that, The trigger optimization time point is used to dynamically adjust the control instruction of the opening and closing of the air release valve of the tongue pressure feedback training instrument, the output frequency of the guide signal is synchronously updated, the stability index in the swallowing training process is determined through real-time sensor data, and the final air release valve opening and closing control strategy of the tongue pressure feedback training instrument and the timing mode of the trigger optimization time point are generated, including: The output frequency of the guide signal is adjusted according to the interval of the trigger optimization time point; the coefficient of variation of the sensor data is calculated as a stability index, the gradient descent method is used to optimize and adjust the parameters based on the stability index, and the air release valve opening and closing control strategy and the timing mode of the trigger optimization time point are generated.

Images