System and method for postoperative recurrent laryngeal nerve function impairment and recovery monitoring and assessment

Abstract

The application discloses a postoperative recurrent laryngeal nerve function injury and recovery monitoring and evaluation system and method, and relates to the technical field of laryngeal nerve function evaluation.The application realizes the transformation of the laryngeal recurrent nerve function state from'subjective observation' to 'objective quantification' by constructing a preoperative individualized neuromuscular vibration fingerprint baseline, converting postoperative neck muscle sound signals into multi-dimensional computable characteristics reflecting nerve discharge rhythm, muscle mechanical modalities and nerve-muscle transmission coupling relationships, and comparing structures under the same stable topology reference.The application can dynamically track the stable maintenance degree of nerve control structure under the condition of non-invasion and continuous wearing, distinguish the nerve source abnormality, muscle source change and normal function recovery trend, significantly reduce the influence of individual anatomic differences, vocalization habits and environmental noise on the evaluation results, and improve the sensitivity, repeatability and long-term trend judgment ability of postoperative function monitoring.

Description

Technical Field

[0001] This invention relates to the field of laryngeal nerve function assessment technology, and in particular to a system and method for monitoring and assessing postoperative recurrent laryngeal nerve function injury and recovery. Background Technology

[0002] The recurrent laryngeal nerve (RLN) is a crucial nerve that innervates the muscles of the larynx, and its damage is a common complication of thoracic and cervical surgeries (such as those for esophageal cancer, thyroid cancer, and lung cancer). Postoperative nerve damage can lead to symptoms such as hoarseness, coughing up water, and difficulty breathing, severely impacting patients' quality of life and potentially delaying subsequent radiotherapy, chemotherapy, or rehabilitation. Therefore, early, continuous, and non-invasive dynamic monitoring of postoperative recurrent laryngeal nerve function is of significant clinical importance.

[0003] Currently, commonly used methods for assessing recurrent laryngeal nerve function include laryngoscopy, electromyography (EMG), acoustic analysis, and percutaneous laryngeal ultrasound. Laryngoscopy allows direct observation of vocal cord movement and is the gold standard for diagnosis, but it is highly invasive, requires specialized personnel, and cannot provide continuous monitoring. EMG records the electrical activity of nerve-innervated muscles, but electrode placement is invasive and the signal is susceptible to interference, limiting its continuous postoperative application. Acoustic analysis and speech assessment can reflect vocal status, but they are greatly affected by speaking habits and background noise, making it difficult to accurately capture nerve function at rest. While percutaneous laryngeal ultrasound is non-invasive and real-time, its sensitivity and specificity are limited by the operator's experience and individual anatomical differences (such as thyroid cartilage calcification). Summary of the Invention

[0004] The purpose of this invention is to provide a system and method for monitoring and evaluating postoperative recurrent laryngeal nerve function injury and recovery, in order to improve the technical problem that the existing technology does not quantify the reference system corresponding to the patient's own preoperative neuromuscular control state, and thus cannot achieve objective, continuous and individualized evaluation of the type of recurrent laryngeal nerve function injury and the degree of recovery.

[0005] To achieve the above-mentioned objectives, the embodiments of the present invention provide the following technical solutions:

[0006] A system for monitoring and assessing postoperative recurrent laryngeal nerve function impairment and recovery, comprising:

[0007] The preoperative acquisition module is used to collect preoperative polymorphic laryngeal mechanical acoustic signals under different physiological states of the patient before surgery through a multi-point array of neck muscle acoustic sensors.

[0008] The time-frequency analysis module is used to perform time-frequency structural analysis on preoperative polymorphic laryngeal mechanosound signals and construct an individual neuromuscular vibration fingerprint baseline model.

[0009] The postoperative acquisition module is used to continuously acquire neck vibration signals at different detection points after the patient's surgery;

[0010] The feature separation module is used to extract the active muscle contraction vibration component, body motion interference component, and respiratory background vibration component of each neck vibration signal using blind source separation and spatial filtering, so as to obtain the postoperative muscle sound signal sequence at each time point.

[0011] The mapping module is used to process the postoperative muscle sound signal sequences through short-time Fourier transform, establish the nerve-muscle-vibration structure mapping relationship, and generate the current neural control feature vector at each time point.

[0012] The index calculation module is used to project the current neural control feature vector and the individual neuromuscular vibration fingerprint baseline model at each time point to calculate the comprehensive stability index at each time point.

[0013] The damage classification module is used to generate damage assessment feature vectors based on the current neural control feature vector and various comprehensive stability maintenance indices, calculate damage deviation indexes, and obtain monitoring and assessment results through preset functional state grading rules.

[0014] Furthermore, the processing procedure of the time-frequency analysis module includes:

[0015] Short-time Fourier transform or continuous wavelet transform were performed on the preoperative polymorphic laryngeal mechanoacoustic signals to obtain the corresponding time-frequency representation matrix:

[0016] Based on the periodicity of the time-frequency representation matrix, the corresponding rhythmic features, modal features, and coupling features are calculated and integrated to generate a multidimensional coupling feature vector.

[0017] The multidimensional coupled feature vectors are mapped to an n-dimensional feature space, and the corresponding stable topological region is used as the baseline model of individual neuromuscular vibration fingerprint.

[0018] Furthermore, the rhythmic characteristics include the coefficient of variation and the pulse group synchronization index; the modal characteristics include energy concentration and frequency drift; and the coupling characteristics include the continuity and phase consistency of the time-frequency energy ridge.

[0019] Furthermore, the separation feature module includes:

[0020] The neck vibration signal was corrected and the cross-correlation matrix, phase difference matrix and spatial covariance matrix between different channels were calculated;

[0021] Based on the cross-correlation matrices, phase difference matrices, and spatial covariance matrices, a linear mixing model is generated and solved using blind source separation to obtain a set of independent source mechanoacoustic components.

[0022] The dominant frequency distribution range, rhythmic period stability, spatial weight distribution and phase propagation direction of each component in the set of independent source mechanosound components are calculated, and the active muscle contraction vibration component, body motion interference component and respiratory background vibration component are determined to generate a postoperative mechanosound signal sequence.

[0023] Furthermore, the calculation of the cross-correlation matrix, phase difference matrix, and spatial covariance matrix between different channels includes:

[0024] The neck vibration signal is corrected to generate a multi-channel vibration signal matrix;

[0025] Calculate the cross-correlation matrix between different channels in a multi-channel vibration signal matrix;

[0026] Calculate the phase difference matrix between different channels in a multi-channel vibration signal matrix;

[0027] Calculate the spatial covariance matrix between different channels in a multi-channel vibration signal matrix.

[0028] Furthermore, the processing procedure of the index calculation module includes:

[0029] Each current neural control feature vector is projected onto the topological space of the individual neuromuscular vibration fingerprint baseline model to generate the corresponding postoperative feature mean vector, postoperative covariance matrix, and normalized postoperative feature vector; the normalized postoperative feature vector includes postoperative rhythm features, postoperative modal features, and postoperative coupling features.

[0030] Based on the normalized postoperative feature vectors and the individual neuromuscular vibration fingerprint baseline model, the corresponding rhythm preservation, modality preservation, and coupling consistency are calculated.

[0031] Based on the retention of each rhythm, the retention of each mode, and the consistency of each coupling, the comprehensive stability retention index at different time points is calculated.

[0032] Furthermore, the processing procedure of the damage classification module includes:

[0033] A multidimensional deviation vector is generated by performing a difference operation on the normalized postoperative feature vector and the mean vector of the individual neuromuscular vibration fingerprint baseline model.

[0034] By merging the comprehensive stability preservation indices and their rhythm preservation, modality preservation, and coupling consistency, a function preservation vector is generated;

[0035] The damage deviation index is calculated based on the multidimensional deviation vector and the function preservation vector.

[0036] The multidimensional deviation vector and the function preservation vector are combined and input into the damage pattern classifier to output the damage-dominant pattern.

[0037] Based on the damage deviation index and damage dominance mode, monitoring and evaluation results are obtained through preset functional state classification rules.

[0038] A method for monitoring and assessing postoperative recurrent laryngeal nerve function impairment and recovery includes:

[0039] The preoperative polymorphic laryngeal mechanical myosal signals under different physiological states of the patient were collected by a multi-point array cervical myosal sensor and the time-frequency structure was analyzed to construct an individual neuromuscular vibration fingerprint baseline model.

[0040] The neck vibration signals of patients were continuously collected at different detection points after surgery, and the active muscle contraction vibration component, body motion interference component and respiratory background vibration component were separated by blind source separation and spatial filtering to obtain the postoperative muscle sound signal sequence at each time point.

[0041] The postoperative myophone signal sequences were processed by short-time Fourier transform to establish a neural-muscle-vibration structure mapping relationship, generate the current neural control feature vector at each time point, and project it in combination with the individual neuromuscular vibration fingerprint baseline model to calculate the comprehensive stability maintenance index at each time point.

[0042] Based on the current neural control feature vector and various comprehensive stability maintenance indices, a damage assessment feature vector is generated, the damage deviation index is calculated, and the monitoring and assessment results are obtained through preset functional state grading rules.

[0043] This invention constructs a preoperative individualized neuromuscular vibration fingerprint baseline for patients, transforming postoperative neck muscle sound signals into multidimensional computable features reflecting nerve discharge rhythms, muscle biomechanical modes, and neuromuscular conduction coupling. Structural comparison is then performed under a stable topological reference, enabling a shift from "subjective observation" to "objective quantification" of the recurrent laryngeal nerve's functional status. This allows for dynamic tracking of the stability of nerve control structures under non-invasive, continuously wearable conditions, distinguishing between neurogenic abnormalities, myogenic changes, and the trend of normal functional recovery. It significantly reduces the impact of individual anatomical differences, vocal habits, and environmental noise on assessment results, improving the sensitivity, repeatability, and long-term trend judgment ability of postoperative functional monitoring. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a system structure diagram in an embodiment of the present invention;

[0046] Figure 2 This is a flowchart of the method in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0048] Please see Figure 1 This embodiment provides a monitoring and assessment system for postoperative recurrent laryngeal nerve function injury and recovery, comprising:

[0049] The preoperative acquisition module is used to collect preoperative polymorphic laryngeal mechanical acoustic signals under different physiological states of the patient before surgery using a multi-point array of neck muscle acoustic sensors. Taking a single patient as an example, in the preoperative healthy state of the patient, laryngeal mechanical acoustic signals are collected at different detection points under various physiological states such as resting breathing, voicing, and coughing using a multi-point array of neck muscle acoustic sensors. All laryngeal mechanical acoustic signals are integrated to generate preoperative polymorphic laryngeal mechanical acoustic signals.

[0050] The time-frequency analysis module is used to perform time-frequency structural analysis on preoperative polymorphic laryngeal mechanosound signals and construct an individual neuromuscular vibration fingerprint baseline model.

[0051] The postoperative data acquisition module is used to continuously collect neck vibration signals from different detection points after surgery. These different detection points include the cricothyroid muscle detection point (reflecting the activity of tone-regulating muscles), the vocal cord closure muscle detection point (reflecting the activity of adductor muscles), and a reference vibration point (located away from the main laryngeal muscles and used for background vibration modeling). Corresponding sensors are set at these different detection points to collect neck vibration signals from the patient at different time points after surgery (on the day of surgery, three days post-surgery, one week post-surgery, and one month post-surgery, etc.).

[0052] The feature separation module is used to extract the active muscle contraction vibration component, body motion interference component, and respiratory background vibration component of each neck vibration signal using blind source separation and spatial filtering, so as to obtain the postoperative muscle sound signal sequence at each time point.

[0053] The mapping module is used to process the postoperative muscle sound signal sequences through short-time Fourier transform, establish the nerve-muscle-vibration structure mapping relationship, and generate the current neural control feature vector at each time point.

[0054] The index calculation module is used to project the current neural control feature vector and the individual neuromuscular vibration fingerprint baseline model at each time point to calculate the comprehensive stability retention index at each time point. The comprehensive stability retention index is obtained by weighted summation of rhythm retention, modality retention, and coupling consistency.

[0055] The damage classification module is used to generate damage assessment feature vectors based on the current neural control feature vector and various comprehensive stability maintenance indices, calculate damage deviation indexes, and obtain monitoring and assessment results through preset functional state grading rules.

[0056] This system also includes a training module for training the damage pattern classifier (Gaussian Mixture Model, GMM). In the training module, the standard Expectation-Maximization (EM) algorithm and unsupervised clustering algorithm are used to train the damage pattern classifier (Gaussian Mixture Model, GMM). Since the standard Expectation-Maximization (EM) algorithm and unsupervised clustering algorithm are existing mature technologies, they will not be described in detail.

[0057] Because each patient exhibits significant differences in neck anatomy, muscle thickness, vocal habits, and tissue damping characteristics, directly using group standards or fixed thresholds may lead to signal bias or misjudgment. Furthermore, traditional methods typically only collect signals under a single state (such as vocalization or resting), failing to comprehensively reflect neuromuscular functional performance under different physiological states, resulting in a lack of sensitivity and precision in functional assessment. Simple amplitude, frequency, or energy comparisons are insufficient to accurately map neural control rhythms, muscle biomechanics, and the integrity of nerve-to-muscle transmission, failing to provide interpretable indicators of functional recovery. Based on these issues, this embodiment designs a time-frequency analysis module. Through time-frequency structure analysis of preoperative multi-state muscle sound signals, it extracts multidimensional features reflecting neural discharge rhythms, laryngeal muscle biomechanical modes, and neuromuscular mechanical coupling, constructing them as a stable topological region in the feature space. This model fully considers individual patient differences and multi-state functional information, providing a reliable reference for postoperative signals, making the calculation of indicators such as rhythm retention, modality retention, and coupling consistency more targeted and accurate, thereby achieving continuous, sensitive, and quantifiable assessment of postoperative neural control function recovery.

[0058] Therefore, the process of the time-frequency analysis module includes:

[0059] S2-1. Perform short-time Fourier transform or continuous wavelet transform on each preoperative polymorphic laryngeal mechanoacoustic signal to obtain the corresponding time-frequency representation matrix:

[0060] S2-2. Based on the periodic characteristics of the time-frequency representation matrix, calculate the corresponding rhythmic features (stability of nerve discharge rhythm), modal features (mechanical state of laryngeal muscles), and coupling features (integrity of nerve-muscle-mechanical transmission) and integrate them to generate a multidimensional coupling feature vector;

[0061] It needs to be explained that, using different detection points as channels, the rhythmic characteristics include the coefficient of variation and the pulse group synchronization index. The peak value of the time-frequency representation matrix is ​​identified, the time interval between consecutive peak values ​​is determined and used as the pulse interval, the ratio of variance to standard deviation within that pulse interval is calculated, and this ratio is used as the coefficient of variation. Finally, according to the formula:

[0062] ;

[0063] Calculate the first The pulse group synchronization index corresponding to each channel ;in, Indicates the number of pulses. Indicates the first The amplitude of each pulse peak, Indicates the average peak amplitude. Represents absolute value. This represents the summation function.

[0064] The lower the coefficient of variation and the higher the pulse group synchronization index, the more stable the patient's neural discharge rhythm.

[0065] The modal characteristics include energy concentration and frequency drift, both of which reflect the stability and mechanical coordination of laryngeal muscle vibration. Energy concentration The corresponding formula is:

[0066] ;

[0067] Frequency drift The corresponding formula is:

[0068] ;

[0069] in, This represents the time-frequency representation result (the first element in the time-frequency representation matrix). The characteristics of each physiological state were obtained through Fourier transform processing. , Representing the frequency and time of the principal vibration mode, respectively. The main vibration mode frequency below, This indicates the upper limit frequency of the analyzed spectrum. This represents the average frequency of the main vibration modes. Indicates frequency, This represents the total number of time points.

[0070] The coupling characteristics include the continuity and phase consistency of the time-frequency energy ridge. Higher continuity and phase consistency indicate a complete nerve-to-muscle mechanical response transmission link and stable vibration mode under neural control. The continuity of the time-frequency energy ridge is the ratio of the continuous ridge duration to the total sampling time. The continuous ridge duration refers to the determination that the main vibration mode remains continuous within the time period when the amplitude of the change in the main vibration frequency corresponding to adjacent moments is less than a preset frequency continuity threshold, as shown in the time-frequency energy distribution map obtained after time-frequency transformation of the muscle sound signal. The total duration obtained by summing all time intervals that meet this continuity determination condition is the continuous ridge duration. This parameter is used to characterize the degree of stability of the main vibration mode within the observation period. The longer the continuous ridge duration, the more stable the muscle vibration mode driven by the nerve and the more complete the nerve-muscle control link; conversely, it reflects frequent interruptions or jumps in the vibration mode, indicating a decrease in neural control stability.

[0071] Phase consistency The corresponding formula is:

[0072] ;

[0073] in, Represents the natural constant. Indicates the first in the preoperative data The first channel and the first Each channel at frequency The phase difference below.

[0074] S2-3. Map the multidimensional coupled feature vectors to the n-dimensional feature space, and use the corresponding stable topological region as the baseline model of individual neuromuscular vibration fingerprint.

[0075] Specifically, the multidimensional coupled feature vectors are standardized to generate standardized multidimensional coupled feature vectors. The mean vector and covariance matrix of the standardized multidimensional coupled feature vectors are then calculated to define the "shape" of the feature distribution under individual health conditions.

[0076] Based on the mean vector and covariance matrix of standardized multidimensional coupled feature vectors, a stable region boundary is constructed using Mahalanobis distance. This stable region boundary is then filtered according to a preset confidence threshold. The resulting stable topological region, obtained by integrating the filtered stable region boundaries, serves as the baseline model for an individual neuromuscular vibration fingerprint. This baseline model is a multidimensional stable structural region representing the overall pattern of neural control, muscle mechanics, and mechanical vibration coupling in the patient's healthy state.

[0077] In this embodiment, Mahalanobis distance is used to define the individual neuromuscular vibration fingerprint baseline model as a high-dimensional ellipsoid. Features falling inside the ellipsoid are considered to be in a "healthy and stable state," while features outside the ellipsoid are considered to be deviating. Individual Neuromuscular Vibration Fingerprint Baseline Model The corresponding formula is:

[0078] ;

[0079] in, This represents the preset confidence threshold. Represents a standardized multidimensional coupled feature vector. This represents the mean of the standardized multidimensional coupled eigenvectors. Represents the covariance matrix. This represents the transpose of the matrix.

[0080] This embodiment analyzes the time-frequency structure of preoperative multi-state muscle sound signals to extract multi-dimensional features such as rhythm stability, vibration modal stability, and neuromuscular mechanical coupling integrity. It also constructs an individual neuromuscular vibration fingerprint baseline model, allowing postoperative assessment to be based on the patient's own structural reference of health status. This effectively eliminates the influence of individual anatomical differences, vocal habits, and tissue damping differences. Combined with multi-point array acquisition, blind source separation, and physiological pattern discrimination techniques, it can accurately separate agonist muscle contraction vibrations from respiratory background and body movement interference, significantly improving the signal-to-noise ratio and feature stability of muscle sound signals. By projecting postoperative feature vectors onto the preoperative stable topological region and calculating the comprehensive stability retention index, it achieves continuous quantification of the degree of deviation in neural control function. This enables earlier identification of nerve conduction abnormalities or muscle mechanical changes, supports recovery trend prediction and damage type differentiation. Compared with traditional assessment methods that rely on subjective judgment or single signal indicators, it has the technical advantages of being non-invasive, continuous, individualized, highly interpretable, and more sensitive to early functional changes.

[0081] Traditional acquisition methods for neck vibration or muscle sound signals typically involve directly bandpass filtering or noise reduction of the raw signal. First, postoperative neck vibration signals simultaneously contain mechanical vibrations generated by active contraction of the laryngeal muscles (truly reflecting neural control), low-frequency periodic vibrations caused by respiratory movements, and interfering vibrations from body movement, swallowing, and skin gliding. Therefore, traditional acquisition methods often result in aliasing of physiological origins in the acquired neck vibration or muscle sound signals. Second, traditional filtering can only process signals by frequency range, but these components overlap in frequency bands and cannot be physiologically separated. This can lead to misinterpreting interference as pathological changes. Changes in respiratory or body movement amplitude can cause frequency drift, altered energy distribution, and fragmented time-frequency structure in the filtered signal, which can be misinterpreted by algorithms as neurological dysfunction. Finally, traditional acquisition methods produce "clean signals" but fail to effectively distinguish whether the vibrations originate from neurally controlled muscle contractions or external movements, thus failing to establish a physiological mapping relationship of "nerve → muscle → mechanical vibration." Therefore, it can be seen that the neck vibration signal is not a single signal, but a superposition field of multiple physiological sources. Thus, the neck vibration signal (mixed vibration field) is decomposed according to physiological sources, changing from the traditional frequency filtering approach to the source separation approach under the constraints of space and physiological mechanisms.

[0082] Furthermore, taking the neck vibration signal at a single time point as an example, the process of separating the feature modules includes:

[0083] S4-1. Correct the neck vibration signal and calculate the cross-correlation matrix, phase difference matrix and spatial covariance matrix between different channels;

[0084] S4-1 includes:

[0085] S4-1-1. Correct the neck vibration signal and generate a multi-channel vibration signal matrix;

[0086] Specifically, different detection points are used as channels, for example, the cricothyroid muscle detection point is channel 1. The neck vibration signal is then sequentially subjected to DC drift correction, sensor amplitude normalization, and contact impedance anomaly detection. High-pass filtering or moving average removal is used to correct DC drift in each channel of the neck vibration signal to eliminate sensor baseline bias. Amplitude normalization is performed on different channel signals based on a reference amplitude or normalization factor to reduce amplitude deviation caused by differences in sensor sensitivity. Simultaneously, anomaly detection of sensor adhesion is performed by combining contact impedance or signal energy distribution characteristics. When abnormally low amplitude, abnormally high noise, or impedance imbalance occurs, it is determined to be poor contact, and the corresponding channel data is removed, generating a multi-channel vibration signal matrix to ensure signal consistency and reliability in subsequent multi-channel spatial analysis.

[0087] S4-1-2, According to the formula:

[0088] ;

[0089] ;

[0090] ;

[0091] Calculate the cross-correlation matrix between different channels in a multi-channel vibration signal matrix. ;in, Represents the time delay variable. Indicates the first The first channel and the first The number of cross-correlation coefficients for each channel This indicates the length of the time window (observation duration) used in a single signal analysis. , Representing time respectively The next Multi-channel vibration signal of each channel, time The next Multi-channel vibration signal with multiple channels. Represents an integral function. Represents the cross-correlation matrix The Middle Line 1 The normalized cross-correlation coefficient of the column. Represents the maximum value function. Indicates the first The cross-correlation coefficient of each channel at zero latency Indicates the first The cross-correlation coefficient of each channel at zero latency Indicates the total number of channels.

[0092] S4-1-3, According to the formula:

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] Calculate the phase difference matrix between different channels in a multi-channel vibration signal matrix. ;in, Indicates frequency Next Fourier signals of each channel, Represents the Fourier function. Indicates the first The first channel and the first Cross-power spectrum of each channel, Indicates the first The complex conjugate of the Fourier signals of each channel Represents a complex phase angle function. Indicates the first The first channel and the first Each channel at frequency The phase difference below.

[0098] S4-1-4, According to the formula:

[0099] ;

[0100] ;

[0101] Calculate the spatial covariance matrix between different channels in a multi-channel vibration signal matrix. ;in, Indicates the first The signal after zero mean of each channel Indicates the first The average signal value of each channel. Indicates the first The signal after zero mean of each channel.

[0102] S4-2. Based on each cross-correlation matrix, each phase difference matrix, and each spatial covariance matrix, a linear mixing model is generated and solved using blind source separation (BSS) to obtain the set of independent source mechanoacoustic components.

[0103] Specifically, based on the spatial propagation characteristics of each cross-correlation matrix, each phase difference matrix, and each spatial covariance matrix, a hybrid matrix is ​​constructed. The cross-correlation matrix reflects the synchronicity of signals in each channel. Channels with high cross-correlation indicate that they may be dominated by the same source signal; therefore, the row values ​​of the corresponding potential physiological vibration sources in the mixing matrix should be similarly distributed across these channels. The phase difference matrix reflects the relative delay information of the signal in the spatial propagation direction. Phase consistency can constrain the phase distribution of the column vectors in the mixing matrix, ensuring that the propagation of the same potential physiological vibration source in different channels conforms to the physical delay law. The spatial covariance matrix characterizes the energy coupling structure of signals in each channel. Principal component or covariance features can guide the amplitude distribution of the mixing matrix, ensuring that the energy projection of each source on the channel is consistent with the actual spatial coupling mode. Furthermore, statistical optimization or constrained least squares methods can be used to process the cross-correlation matrices, phase difference matrices, and spatial covariance matrices to calculate the mixing matrix. This facilitates the subsequent separation of agonist contraction, respiratory background, and somatic disturbances, while ensuring the physical and physiological interpretability of each component.

[0104] Then, according to the formula:

[0105] ;

[0106] in, Represents the potential physiological vibration source vector. This represents the observed signal matrix.

[0107] In the solution process, the postoperative neck vibration signals acquired by the multi-point array are regarded as a linear superposition of multiple physiological vibration sources, forming a linear mixture model. Using the blind source separation (BSS) method, by assuming that the signals of each source are statistically independent or non-Gaussian distributed, the observed signal matrix is ​​solved. The mixture matrix is ​​estimated by statistical optimization or constrained least squares method, and each independent potential physiological vibration source is inversely obtained to generate a set of independent source mechanomotor components.

[0108] S4-3. Calculate the dominant frequency distribution range, rhythmic period stability, spatial weight distribution and phase propagation direction of each component in the independent source mechanoacoustic component set, and determine the agonist muscle contraction vibration component, body motion interference component and respiratory background vibration component to generate a postoperative muscle sound signal sequence.

[0109] It needs to be explained that, in terms of a component For example, the power spectrum of this component is obtained by Fast Fourier Transform (FFT) and the cumulative energy distribution is calculated. The frequencies corresponding to the cumulative energy distribution reaching 2.5% and 97.5% are set as the main frequency distribution range, which is used to distinguish between low-frequency respiratory vibration, muscle contraction vibration and high-frequency body motion interference.

[0110] Calculate the time difference (interpulse period) between consecutive pulse peaks of the component, calculate the variance and standard deviation within the interpulse period, and use the ratio between the variance and standard deviation as the rhythm period stability. Components with lower rhythm period stability and periodic peak groups correspond to muscle contraction driven by neural control, while components with higher rhythm period stability and irregularity correspond to body movement disturbances or random vibrations.

[0111] The first of the mixing matrix The column coefficients represent the projection coefficients of the component on each channel, and the sequence of projection coefficients in each column represents the spatial weight distribution. For example, if the detection points include the cricothyroid muscle, vocal cord closure muscle detection points, reference vibration points, or points far from the muscle, and are numbered, then the corresponding projection coefficient sequence is: . to Represents the projection coefficients at different detection points, when to If the value of the component is greater than the other two projection coefficients and the values ​​of the other two projection coefficients are close to 0, it indicates that the physiological vibration source corresponding to this component mainly has its energy concentrated in the muscle channel and belongs to the agonist muscle contraction vibration component.

[0112] By calculating the phase difference of this component in the frequency domain of different channels, if the phase changes slowly and linearly along the longitudinal neck channel (phase propagation direction) and low frequency is dominant, it indicates respiratory background vibration; if the local phase is consistent and the frequency is concentrated in the muscle vibration bandwidth, it is the agonist muscle contraction vibration component; if the phase is random and irregular, it is body motion interference.

[0113] Because the respiratory background component is a low-frequency, globally phase-progressive, spatially widespread, and rhythmically stable component; the agonist muscle contraction vibration component is a mid-frequency, locally spatially concentrated, and rhythmically stable or regular component; and the body motion interference component is a broadband, high-amplitude burst, spatially irregular, and rhythmically unstable component, a set of independent source mechanomotor sound components is filtered based on these characteristics and preset thresholds or ranges. Body motion interference components and respiratory background vibration components are removed, leaving only the agonist muscle contraction vibration component, thus generating a postoperative muscle sound signal sequence. Since selecting data using preset thresholds and ranges is a standard setting, it will not be described in detail here.

[0114] Repeat steps S4-1 to S4-3 to obtain postoperative muscle sound signal sequences at all time points.

[0115] This embodiment utilizes blind source separation and physiological pattern discrimination of multi-point array muscle sound signals. This step can accurately decompose the postoperative signal into agonist muscle contraction, respiratory background, and body movement interference components. The rhythmic stability, dominant frequency mode, and spatial concentration of agonist muscle contraction provide a highly sensitive representation of neural control and muscle mechanics, while eliminating respiratory and body movement interference, effectively improving the signal-to-noise ratio. This enables continuous, non-invasive, and real-time monitoring of the cervical recurrent laryngeal nerve-innervated muscle groups, providing a reliable foundation for subsequent time-frequency structural analysis and functional assessment.

[0116] The mapping module processes the postoperative muscle sound signal sequences using the same method as S2-2, establishes a neural-muscle-vibration structure mapping relationship, and generates current neural control feature vectors at different time points. This is beneficial for projecting the current neural control feature vectors onto the topological space of the individual neuromuscular vibration fingerprint baseline model.

[0117] The processing steps of the index calculation module include:

[0118] S6-1. Using the same method as S2-3, process the current neural control feature vectors at different time points, and project each current neural control feature vector onto the topological space of the individual neuromuscular vibration fingerprint baseline model to generate the corresponding postoperative feature mean vector, postoperative covariance matrix, and normalized postoperative feature vector; the normalized postoperative feature vector includes postoperative rhythm features, postoperative modal features, and postoperative coupling features.

[0119] Postoperative feature mean vector Postoperative covariance matrix and normalized postoperative feature vector The relationship between them satisfies the following conditions:

[0120] ;

[0121] in, It represents the inverse square root of the postoperative covariance matrix (used to normalize different dimensions and correlations). This represents the current neural control feature vector.

[0122] S6-2. Based on the normalized postoperative feature vectors and the individual neuromuscular vibration fingerprint baseline model, calculate the corresponding rhythm retention, modality retention and coupling consistency.

[0123] It is important to explain that, due to the inherent hierarchical structure of the neuromuscular control pathway—where neural firing rhythms first determine the driving mode, then form the mechanical modes of muscle contraction, ultimately manifesting as measurable mechanical vibration structures—calculating only the overall Mahalanobis distance between the complete feature vector and the preoperative mean as an indicator of postoperative status, while reflecting the overall degree of deviation, fails to distinguish whether the deviation originates from neural rhythm disturbances, muscle mode changes, or abnormalities in the neuromuscular conduction pathway, thus lacking the ability to explain physiological mechanisms. Therefore, it is necessary to model the features in blocks according to structural hierarchy, calculating the retention degree separately, so that each type of retention degree corresponds to a specific physiological link. Furthermore, since there are statistical correlations among the features—for example, changes in dominant frequency are often accompanied by changes in energy distribution, and changes in coherence may affect phase stability—using simple Euclidean distance would ignore the covariance structure, potentially leading to overestimation or underestimation of the degree of deviation. Therefore, Mahalanobis distance is used to introduce a covariance matrix, achieving a scale-independent deviation measure while considering feature correlations. Furthermore, preoperative health status is not a single ideal point in a statistical sense, but rather forms a stable topological region with a certain variance range in a multidimensional feature space. Therefore, the definition of retention should reflect the relative position of the current feature within this stable region, rather than a simple comparison with a fixed value. Through hierarchical modeling, covariance constraints, and stable region normalization, retention can be made both statistically rigorous and based on a clear physiological interpretation. Therefore, this embodiment sets rhythm retention, modality retention, and coupling consistency.

[0124] Taking a specific time point as an example, rhythm retention is used to characterize the degree to which the postoperative neural discharge rhythm characteristics are maintained relative to the preoperative healthy rhythm structure. It can measure whether "the neural drive rhythm is still stable, orderly, and close to the preoperative state." The corresponding formula is:

[0125] ;

[0126] ;

[0127] in, Indicates the Mahalanobis distance of the rhythm. This represents the postoperative rhythm feature vector within the normalized postoperative feature vector at that time point. This represents the preoperative rhythm mean vector of the individual neuromuscular vibration fingerprint baseline model. This represents the preoperative rhythm covariance matrix of the baseline model of an individual's neuromuscular vibration fingerprint. Indicates the degree of rhythm maintenance. This represents an exponential function with base e. This represents the threshold of the rhythm stability region.

[0128] Modal retention is used to characterize the degree to which the postoperative laryngeal muscle mechanical vibration modal structure is retained relative to the preoperative healthy vibration mode. It measures whether "the dominant vibration mode formed by muscle contraction still retains its original structure." The corresponding formula is:

[0129] ;

[0130] ;

[0131] in, Represents the modal Mahalanobis distance. This represents the postoperative modal feature vector within the normalized postoperative feature vector at that time point. This represents the preoperative modal mean vector of the baseline model of an individual's neuromuscular vibration fingerprint. This represents the preoperative modal covariance matrix of the baseline model of an individual's neuromuscular vibration fingerprint. Indicates modality retention. This represents the modal stability threshold.

[0132] Coupling consistency is used to characterize the degree to which the postoperative neuromuscular-mechanical vibration transmission relationship is maintained relative to the preoperative coupling structure. It can measure whether "the synchronicity and structural integrity between neural drive and mechanical vibration still exist." The corresponding formula is:

[0133] ;

[0134] ;

[0135] in, Represents the coupling Mahalanobis distance. This represents the postoperative coupling feature vector within the normalized postoperative feature vector at that time point. This represents the preoperative coupling covariance matrix of the baseline model of an individual neuromuscular vibration fingerprint. This represents the preoperative coupling mean vector of the baseline model of an individual neuromuscular vibration fingerprint. Indicates coupling consistency. This represents the coupling stability threshold.

[0136] S6-3. Based on the rhythm retention, modality retention, and coupling consistency of each rhythm, calculate the comprehensive stability retention index at different time points. The corresponding formula is:

[0137] ;

[0138] in, , , These represent rhythm weights, modal weights, and coupling weights, respectively. and .

[0139] This embodiment projects the current postoperative neural control feature vector onto a stable topological region formed by an individual baseline model constructed from preoperative multi-state health data. It then calculates rhythm retention, modality retention, and neuromuscular coupling consistency, fusing them to obtain a comprehensive stability retention index. This ensures that postoperative assessment is based on the patient's own healthy neuromuscular control structure, rather than the group average threshold, effectively eliminating the influence of individual anatomical differences, tissue mechanics differences, and vocal habits. Simultaneously, by characterizing the structural stability of the three physiological links—neural drive, muscle vibration mode, and conduction pathway—through hierarchical retention, the monitoring results not only reflect the overall functional change amplitude but also indicate the physiological level of the abnormality source. This improves the sensitivity and interpretability of identifying early subtle functional decline, making it suitable for continuous dynamic trend assessment and quantitative analysis of the recovery process after surgery.

[0140] The processing steps of the damage classification module include:

[0141] S7-1. Perform a difference operation on the normalized postoperative feature vector and the mean vector of the individual neuromuscular vibration fingerprint baseline model to generate a multidimensional deviation vector; the mean vector of the individual neuromuscular vibration fingerprint baseline model refers to the mean of the preoperative multidimensional coupled feature vector.

[0142] S7-2. Merge each comprehensive stability preservation index and its rhythm preservation degree, modality preservation degree and coupling consistency to generate a function preservation vector;

[0143] S7-3. Based on multidimensional deviation vectors and function preservation vectors, calculate the damage deviation index using weighted Mahalanobis distance. The corresponding formula is:

[0144] ;

[0145] ;

[0146] in, This represents the damage assessment feature vector obtained by merging the multidimensional deviation vector and the function preservation vector. , These represent preoperative data from healthy individuals (or data from multiple preoperative time periods for individual patients) at different points in time before surgery. The mean vector and the inverse covariance matrix in space, Represents the normalized scaling parameter. This represents the initial damage deviation index.

[0147] S7-4. Merge the multidimensional deviation vector and the function preservation vector and input them into the damage pattern classifier (Gaussian mixture model, GMM) to output the damage-dominant pattern.

[0148] Specifically, this embodiment employs a Gaussian Mixture Model (GMM) to construct an injury pattern classifier. Under unsupervised conditions, the model automatically divides the feature space into several high-dimensional ellipsoidal clusters, each corresponding to a typical functional abnormality pattern. By analyzing and clinically validating the composition of the central feature vectors of each cluster, a physiological interpretation label can be assigned to each cluster. For example: rhythm instability type, characterized by a significant decrease in rhythm retention and a predominance of deviation components related to neural discharge rhythms, suggesting impaired neural conduction function; modal drift type, characterized by a significant decrease in modal retention and a predominance of deviation components related to laryngeal muscle biomechanical modes, suggesting myogenic changes or compensatory contraction patterns; coupling mismatch type, characterized by a significant decrease in coupling consistency and a predominance of deviation components related to the integrity of neuromuscular mechanical conduction, suggesting neuromuscular junction or mechanical conduction link disorders.

[0149] In actual monitoring, the damage assessment feature vector is input into a trained Gaussian mixture model, and the posterior probability of the feature vector belonging to each preset mode cluster is calculated. The mode with the highest posterior probability is taken as the current dominant damage mode, and this maximum probability value is output as the confidence score. If the confidence score is lower than a preset threshold (e.g., 0.6), the current state is determined to be a mixed mode or an uncertain mode, suggesting that multiple physiological mechanisms may be involved, requiring further differentiation based on other clinical information. Conversely, based on the characteristics of each cluster, the dominant damage mode is determined to be rhythm instability type / modal drift type / coupling mismatch type.

[0150] S7-5. Based on the damage deviation index and damage dominance mode, the monitoring and evaluation results are obtained through preset functional state classification rules.

[0151] Specifically, based on the damage deviation index and referring to a pre-defined five-level functional status grading table (Level I to Level V), the current functional status level is determined. The current functional status level and the dominant damage pattern are then integrated to generate monitoring and assessment results.

[0152] For example, the preset five-level functional status classification table is as follows: if the damage deviation index is less than 0.2, the current functional status level is Level I, representing a healthy and stable state; if the damage deviation index is between 0.2 and 0.4, the current functional status level is Level II, representing a mild deviation; if the damage deviation index is between 0.4 and 0.6, the current functional status level is Level III, representing a moderate deviation; if the damage deviation index is between 0.6 and 0.8, the current functional status level is Level IV, representing a severe deviation; if the damage deviation index is greater than or equal to 0.8, the current functional status level is Level V, representing a serious deviation.

[0153] Please see Figure 2This embodiment provides a method for monitoring and evaluating postoperative recurrent laryngeal nerve function damage and recovery. Figure 2 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not impose any limitations on this.

[0154] A method for monitoring and assessing postoperative recurrent laryngeal nerve function impairment and recovery includes:

[0155] S1. Collect preoperative polymorphic laryngeal mechanical myosal signals under different physiological states of the patient before surgery using a multi-point array cervical myosal sensor and perform time-frequency structure analysis to construct an individual neuromuscular vibration fingerprint baseline model.

[0156] S2. Continuously collect neck vibration signals at different detection points after the patient's surgery and use blind source separation and spatial filtering to separate the active muscle contraction vibration component, body motion interference component and respiratory background vibration component to obtain the postoperative muscle sound signal sequence at each time point.

[0157] S3. Process the postoperative muscle sound signal sequences through short-time Fourier transform, establish the nerve-muscle-vibration structure mapping relationship, generate the current neural control feature vector at each time point, project it in combination with the individual neuromuscular vibration fingerprint baseline model, and calculate the comprehensive stability maintenance index at each time point.

[0158] S4. Based on the current neural control feature vector and various comprehensive stability maintenance indices, generate a damage assessment feature vector, calculate the damage deviation index, and obtain the monitoring and assessment results through preset functional state grading rules.

[0159] In summary, this invention constructs a preoperative individualized neuromuscular vibration fingerprint baseline for patients, transforming postoperative neck muscle sound signals into multidimensional computable features reflecting nerve discharge rhythms, muscle biomechanical modes, and neuromuscular conduction coupling. Structural comparison is then performed under a stable topological reference, enabling a shift from "subjective observation" to "objective quantification" of the recurrent laryngeal nerve functional status. This allows for dynamic tracking of the stability of nerve control structures under non-invasive, continuously wearable conditions, distinguishing between neurogenic abnormalities, myogenic changes, and the trend of normal functional recovery. It significantly reduces the impact of individual anatomical differences, vocal habits, and environmental noise on assessment results, improving the sensitivity, repeatability, and long-term trend judgment ability of postoperative functional monitoring.

[0160] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0161] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0162] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A system for monitoring and assessing postoperative recurrent laryngeal nerve function injury and recovery, characterized in that, include: The preoperative acquisition module is used to collect preoperative polymorphic laryngeal mechanical acoustic signals under different physiological states of the patient before surgery through a multi-point array of neck muscle acoustic sensors. The time-frequency analysis module is used to perform time-frequency structural analysis on preoperative polymorphic laryngeal mechanosound signals and construct an individual neuromuscular vibration fingerprint baseline model. The postoperative acquisition module is used to continuously acquire neck vibration signals at different detection points after the patient's surgery; The feature separation module is used to extract the active muscle contraction vibration component, body motion interference component, and respiratory background vibration component of each neck vibration signal using blind source separation and spatial filtering, so as to obtain the postoperative muscle sound signal sequence at each time point. The mapping module is used to process the postoperative muscle sound signal sequences through short-time Fourier transform, establish the nerve-muscle-vibration structure mapping relationship, and generate the current neural control feature vector at each time point. The index calculation module is used to project the current neural control feature vector and the individual neuromuscular vibration fingerprint baseline model at each time point to calculate the comprehensive stability index at each time point. The damage classification module is used to generate damage assessment feature vectors based on the current neural control feature vector and various comprehensive stability maintenance indices, calculate damage deviation index, and obtain monitoring and assessment results through preset functional state grading rules. The processing procedure of the time-frequency analysis module includes: Short-time Fourier transform or continuous wavelet transform were performed on the preoperative polymorphic laryngeal mechanoacoustic signals to obtain the corresponding time-frequency representation matrix: Based on the periodicity of the time-frequency representation matrix, the corresponding rhythmic features, modal features, and coupling features are calculated and integrated to generate a multidimensional coupling feature vector. The multidimensional coupled feature vectors are mapped to an n-dimensional feature space, and the corresponding stable topological region is used as the baseline model of individual neuromuscular vibration fingerprint. The processing procedure of the index calculation module includes: Each current neural control feature vector is projected onto the topological space of the individual neuromuscular vibration fingerprint baseline model to generate the corresponding postoperative feature mean vector, postoperative covariance matrix, and normalized postoperative feature vector; the normalized postoperative feature vector includes postoperative rhythm features, postoperative modal features, and postoperative coupling features. Based on the normalized postoperative feature vectors and the individual neuromuscular vibration fingerprint baseline model, the corresponding rhythm preservation, modality preservation, and coupling consistency are calculated. Based on the retention of each rhythm, the retention of each mode, and the consistency of each coupling, the comprehensive stability retention index at different time points is calculated.

2. The postoperative recurrent laryngeal nerve function injury and recovery monitoring and assessment system according to claim 1, characterized in that, The rhythmic features include the coefficient of variation and the pulse group synchronization index; the modal features include energy concentration and frequency drift; and the coupling features include the continuity and phase consistency of the time-frequency energy ridge.

3. The postoperative recurrent laryngeal nerve function injury and recovery monitoring and assessment system according to claim 1, characterized in that, The separation feature module includes: The neck vibration signal was corrected and the cross-correlation matrix, phase difference matrix and spatial covariance matrix between different channels were calculated. Based on the cross-correlation matrices, phase difference matrices, and spatial covariance matrices, a linear mixing model is generated and solved using blind source separation to obtain a set of independent source mechanoacoustic components. The dominant frequency distribution range, rhythmic period stability, spatial weight distribution and phase propagation direction of each component in the set of independent source mechanosound components are calculated, and the active muscle contraction vibration component, body motion interference component and respiratory background vibration component are determined to generate a postoperative mechanosound signal sequence.

4. The postoperative recurrent laryngeal nerve function injury and recovery monitoring and assessment system according to claim 3, characterized in that, The calculation of the cross-correlation matrix, phase difference matrix, and spatial covariance matrix between different channels includes: The neck vibration signal is corrected to generate a multi-channel vibration signal matrix; Calculate the cross-correlation matrix between different channels in a multi-channel vibration signal matrix; Calculate the phase difference matrix between different channels in a multi-channel vibration signal matrix; Calculate the spatial covariance matrix between different channels in a multi-channel vibration signal matrix.

5. The postoperative recurrent laryngeal nerve function injury and recovery monitoring and assessment system according to claim 1, characterized in that, The processing steps of the damage classification module include: A multidimensional deviation vector is generated by performing a difference operation on the normalized postoperative feature vector and the mean vector of the individual neuromuscular vibration fingerprint baseline model. By merging the comprehensive stability preservation indices and their rhythm preservation, modality preservation, and coupling consistency, a function preservation vector is generated; The damage deviation index is calculated based on the multidimensional deviation vector and the function preservation vector. The multidimensional deviation vector and the function preservation vector are combined and input into the damage pattern classifier to output the damage-dominant pattern. Based on the damage deviation index and damage dominance mode, monitoring and evaluation results are obtained through preset functional state classification rules.

6. A method for monitoring and assessing postoperative recurrent laryngeal nerve function injury and recovery, used to implement the postoperative recurrent laryngeal nerve function injury and recovery monitoring and assessment system as described in any one of claims 1 to 5, characterized in that, include: The preoperative polymorphic laryngeal mechanical myosal signals under different physiological states of the patient were collected by a multi-point array cervical myosal sensor and the time-frequency structure was analyzed to construct an individual neuromuscular vibration fingerprint baseline model. The neck vibration signals of patients were continuously collected at different detection points after surgery, and the active muscle contraction vibration component, body motion interference component and respiratory background vibration component were separated by blind source separation and spatial filtering to obtain the postoperative muscle sound signal sequence at each time point. The postoperative myophone signal sequences were processed by short-time Fourier transform to establish a neural-muscle-vibration structure mapping relationship, generate the current neural control feature vector at each time point, and project it in combination with the individual neuromuscular vibration fingerprint baseline model to calculate the comprehensive stability maintenance index at each time point. Based on the current neural control feature vector and various comprehensive stability maintenance indices, a damage assessment feature vector is generated, the damage deviation index is calculated, and the monitoring and assessment results are obtained through preset functional state grading rules.

Images