Non-invasive obstructive sleep apnea awakening cause assessment method and non-invasive obstructive sleep apnea awakening cause assessment system

Abstract

The invention discloses a non-invasive obstructive sleep apnea awareness cause assessment method and a non-invasive obstructive sleep apnea awareness cause assessment system. The method comprises the following steps: acquiring an end position of each OSA event of an OSA patient; acquiring a lower jaw myoelectricity signal and an electroencephalogram signal of the OSA patient, and based on the ending position of each OSA event, calculating the proportion of the lower jaw myoelectricity change prior to the electroencephalogram change in the ending period of all the OSA events as an awakening pathogenesis index of the OSA patient; and obtaining anatomical clinical parameters, PSG parameters, CT parameters and awakening cause indexes of an OSA patient, and inputting the obtained parameters into a preset OSA operation curative effect prediction model to output a postoperative AHI prediction result of the patient. The invention provides an awakening disease cause index capable of quantifying muscle functions and an awakening threshold value at the same time, the awakening disease cause index is used for establishing a subtype suitable for an operation, an OSA operation curative effect prediction model is constructed in combination with anatomy related parameters, and the operation effect can be effectively predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to a non-invasive obstructive sleep apnea arousal cause evaluation method, and also relates to a corresponding non-invasive obstructive sleep apnea arousal cause evaluation system, and belongs to the technical field of medical and health information science. BACKGROUND

[0002] Obstructive sleep apnea (OSA) is a disease in which the upper airway is partially or completely obstructed during sleep. It has been proven to be an independent risk factor for cardiovascular and cerebrovascular diseases, metabolic disorders and other systemic diseases. Since it is a heterogeneous disease caused by multiple factors, its etiology spectrum at least covers anatomical abnormalities, high loop gain, low muscle response and low arousal threshold, therefore, it is crucial to determine whether the low arousal threshold dominates the individual etiology for selecting the appropriate treatment plan. However, the current methods for evaluating the low arousal threshold generally rely on invasive procedures, or have large devices and complex processes, which are significantly limited in clinical application, and there is an urgent need to develop a portable, non-invasive and easy-to-promote arousal cause evaluation method within the framework of routine sleep monitoring.

[0003] The termination of each respiratory event is inseparable from two competing mechanisms: one is the passive activation of airway dilator muscles when chemical or negative pressure stimulation exceeds the effective recruitment threshold of muscles, which directly opens the airway; the other is the awakening of the cerebral cortex when the stimulation exceeds the arousal threshold, which indirectly restores ventilation through central command. For individuals with low arousal threshold, the second mechanism often dominates, showing that arousal occurs earlier than or even completely suppresses the effective contraction of muscles, resulting in the phenomenon of "arousal first, muscle response later" or even "only arousal, no muscle response". Therefore, as long as the sequence of arousal and muscle activation can be captured non-invasively, the weight of arousal threshold in the etiology can be quantified.

[0004] Electroencephalogram (EEG) can reflect the activity of cerebral cortex in real time, and is recognized as an arousal determination signal; electromyogram (EMG) can record the contraction activity of the masseter muscle group, which is one of the airway dilator muscles. By comparing the sequence of changes before and after the end of the same respiratory event, key information can be obtained: if the change in masseter EMG is earlier than the change in EEG, it indicates that the patient's muscle function is good and the arousal threshold is high, and routine surgery may bring significant benefits; on the contrary, if the change in EEG is earlier than the change in masseter EMG, it indicates that the arousal threshold is low and the muscle compensation is insufficient, and the effect of surgery may be limited. Therefore, there is an urgent need in the industry to develop a simple and easy-to-use and completely non-invasive "arousal cause index", which is the proportion of events in which "the change in masseter EMG precedes the change in EEG" during the end of all respiratory events of the patient, to guide clinical decision-making and deepen the understanding of the pathophysiological mechanism of OSA. SUMMARY

[0005] The primary technical problem to be solved by the present application is to provide a non-invasive obstructive sleep apnea arousal cause evaluation method.

[0006] Another technical problem to be solved by the present application is to provide a non-invasive obstructive sleep apnea arousal cause evaluation system.

[0007] To achieve the above technical purposes, the present application adopts the following technical solutions: According to a first aspect of an embodiment of the present application, a non-invasive obstructive sleep apnea arousal cause evaluation method is provided, comprising the following steps: obtaining the end position of each OSA event of an OSA patient; obtaining the myoelectric signal of the lower jaw and the electroencephalogram signal of the OSA patient, and based on the end position of each OSA event, calculating the proportion of the change of the myoelectric signal of the lower jaw preceding the change of the electroencephalogram signal during the end period of all OSA events as the arousal cause index of the OSA patient; obtaining the clinical parameters, PSG parameters and CT parameters of the OSA patient based on anatomy, wherein the clinical parameters at least include the body mass index, the PSG parameters at least include the apnea-hypoapnea index, the arousal index and the time occupied by the blood oxygen being lower than 90%, and the CT parameters at least include the vertical distance between the lower edge of the mandible and the lower edge of the hyoid bone and the airway length; inputting the clinical parameters, PSG parameters, CT parameters and the arousal cause index into a preset OSA surgical efficacy prediction model to output the postoperative apnea-hypoapnea index prediction result of the OSA patient; wherein a data set is formed based on the clinical parameters, CT parameters, PSG parameters and arousal cause index of a plurality of historical patients, and the data set is input into a plurality of machine learning models for model training, and by comparing the prediction performance of the plurality of machine learning models, the machine learning model with the highest prediction performance is selected as the OSA surgical efficacy prediction model.

[0008] Preferably, the arousal cause index is obtained by the following steps: According to the end position of each OSA event, the first transfer entropy from the myoelectric signal of the lower jaw to the electroencephalogram signal and the second transfer entropy from the electroencephalogram signal to the myoelectric signal of the lower jaw are calculated when the time difference is 1 second during the end period of the OSA event; If the first transfer entropy is greater than the second transfer entropy, it indicates that the change of the myoelectric signal of the lower jaw precedes the change of the electroencephalogram signal in the OSA event; otherwise, it indicates that the change of the electroencephalogram signal precedes the change of the myoelectric signal of the lower jaw; statistically counting the number of OSA events in which the change of the myoelectric signal of the lower jaw precedes the change of the electroencephalogram signal, denoted as m; m / n is taken as the arousal cause index of the OSA patient, where n is the total number of OSA events identified based on the OSA patient's mandibular myoelectric signal.

[0009] Preferably, the non-invasive obstructive sleep apnea arousal cause evaluation method further comprises: obtaining the arousal cause index of the historical patient who has undergone surgery and the surgery effectiveness data; Based on the arousal cause index and the surgery effectiveness data of the historical patient who has undergone surgery, the ROC curve for predicting the surgery effectiveness based on the arousal cause index is plotted with the specificity as the horizontal coordinate and the sensitivity as the vertical coordinate; Calculate the area under the ROC curve, and determine the surgery subtype discrimination value according to the maximum value of the sum of sensitivity and specificity; The arousal cause index of the OSA patient is compared with the surgery subtype discrimination value. If the arousal cause index of the OSA patient is greater than the surgery subtype discrimination value, the OSA patient is suitable for surgery; otherwise, the OSA patient is not suitable for surgery.

[0010] Preferably, the maximum value of the sum of sensitivity and specificity under the ROC curve is 0.78.

[0011] Preferably, the OSA surgery effectiveness prediction model is constructed by the following steps: Based on the surgery data of the historical patient, select the parameters in the surgery data that have a correlation with the postoperative apnea-hypoventilation index greater than a first threshold, and eliminate the parameters that have a correlation greater than a second threshold, as the anatomical input parameters; wherein the anatomical input parameters at least include clinical parameters, PSG parameters and CT parameters.

[0012] Obtain the arousal cause index of the historical patient to form a data set with the anatomical input parameters, and divide the data set into a training set and a test set; Input the training set into a plurality of machine learning models for model training, and test each machine learning model through the test set; The regression performance of each machine learning model after model testing is evaluated by using Pearson correlation coefficient, root mean square error and mean absolute error, and the classification performance of each machine learning model after model testing is evaluated by using accuracy, sensitivity and specificity; Based on the regression performance and classification performance of each machine learning model, select the machine learning model with the highest comprehensive prediction performance as the OSA surgery effectiveness prediction model.

[0013] Preferably, the machine learning model comprises one or more of an artificial neural network model, a support vector regression model, a k-nearest neighbor model, a random forest model, an XGBoost model, and a linear regression model.

[0014] Preferably, the arousal cause indicator is negatively correlated with the PSG parameters, the body mass index, and the age, and is positively correlated with the arousal threshold.

[0015] According to a second aspect of the embodiments of the present application, a non-invasive obstructive sleep apnea arousal cause evaluation system is provided, comprising: a data acquisition unit configured to acquire a lower jaw myoelectric signal, an electroencephalogram signal, an anatomic-based clinical parameter, a PSG parameter, and a CT parameter of an OSA patient; a myoelectric recognition unit preconfigured with an OSA recognition model and connected to the data acquisition unit, configured to recognize the lower jaw myoelectric signal, thereby acquiring an ending position of each OSA event of the OSA patient; a calculation unit connected to the myoelectric recognition unit and the data acquisition unit, configured to calculate a proportion of a change in the lower jaw myoelectric signal preceding a change in the electroencephalogram signal during the ending of all OSA events as an arousal cause indicator of the OSA patient based on the ending position of each OSA event and the electroencephalogram signal of the OSA patient; an output unit preconfigured with an OSA surgical efficacy prediction model and connected to the calculation unit and the data acquisition unit, configured to input the clinical parameter, the CT parameter, the PSG parameter, and the arousal cause indicator into the OSA surgical efficacy prediction model, thereby outputting a postoperative apnea-hypopnea index prediction result of the OSA patient; Preferably, a data set is formed based on the clinical parameters, the CT parameters, the PSG parameters, and the arousal cause indicators of a plurality of historical patients, and the data set is input into a plurality of machine learning models for model training. The machine learning model with the highest prediction performance is selected as the OSA surgical efficacy prediction model by comparing the prediction performances of the plurality of machine learning models.

[0016] Preferably, the OSA recognition model is formed by joint training of an LSTM model and a CRF model.

[0017] According to a third aspect of the embodiments of the present application, a non-invasive obstructive sleep apnea arousal cause evaluation system is provided, comprising a processor and a memory. The processor reads a computer program in the memory, and is configured to implement the non-invasive obstructive sleep apnea arousal cause evaluation method.

[0018] Compared with the prior art, the present application proposes an arousal cause index that can simultaneously quantify muscle function and arousal threshold based on the analysis of the sequence of changes in the electromyogram of the mandible and the electroencephalogram during the end of a respiratory event, which is defined as the proportion of the OSA events of the patient in which the change in the electromyogram of the mandible precedes the change in the electroencephalogram during the end of the event. The arousal cause index is combined with surgical effectiveness data to establish a suitable subtype for surgery, which helps to guide the clinical evaluation scheme and provides a new perspective for elucidating the physiological and pathological mechanisms of OSA. On this basis, the present application combines the arousal cause index with clinical parameters of anatomy, PSG parameters and CT parameters, thereby training multiple machine learning models to propose an OSA surgical efficacy prediction model based on anatomical factors and the arousal cause index, which helps to screen patients suitable for surgical treatment and assists doctors in providing personalized treatment for patients. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 For the present application, the influence of the changes in the electroencephalogram and the electromyogram of the mandible on surgery is shown in the schematic diagram. Figure 2 The overall flowchart of the non-invasive obstructive sleep apnea arousal cause evaluation method provided by the first embodiment of the present application is shown in the schematic diagram. Figure 3A For the present application, the correlation between the arousal cause index and age is shown in the schematic diagram. Figure 3B For the present application, the correlation between the arousal cause index and the severity of OSA is shown in the schematic diagram. Figure 3C For the present application, the correlation between the arousal cause index and the minimum esophageal pressure is shown in the schematic diagram. Figure 4 For the present application, the construction principle of the OSA surgical efficacy prediction model in the first embodiment is shown in the schematic diagram. Figure 5 For the present application, the scatter plot of the postoperative AHI predicted value and the true value based on the model of anatomical factors and the arousal cause index in the first embodiment is shown in the schematic diagram. Figure 6 For the present application, the histogram of the difference between the postoperative AHI predicted value and the true value based on the model of anatomical factors and the arousal cause index in the first embodiment is shown in the schematic diagram. Figure 7 For the present application, the importance of each feature of the XGBoost model in the first embodiment is shown in the schematic diagram. Figure 8 The overall architecture of the OSA precise evaluation software provided by the first embodiment of the present application is shown in the schematic diagram. Figure 9 The user interface for OSA diagnosis and respiratory event recognition in the OSA precise evaluation software provided by the first embodiment of the present application is shown in the schematic diagram. Figure 10 A schematic diagram of a surgical effect prediction user interface provided in the OSA precise evaluation software of the first embodiment of the present application; Figure 11 A structural schematic diagram of a non-invasive obstructive sleep apnea arousal cause evaluation system provided in the second embodiment of the present application; Figure 12 A structural schematic diagram of another non-invasive obstructive sleep apnea arousal cause evaluation system provided in the third embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical content of the present application will be described in detail below in combination with the drawings and specific embodiments.

[0021] There are two mechanisms for arousal cause to resume respiratory events: the first mechanism is that the muscle opens the airway through self-regulation, which is manifested as muscle activation preceding arousal or not accompanied by arousal during the end of the respiratory event. The second mechanism is that when the self-regulation of the muscle cannot end the respiratory event, the brain cortex arousal needs to be involved in the regulation. At this time, the termination of the respiratory event is mainly controlled by arousal, and arousal precedes muscle activation. Based on the above two mechanisms, as shown in the table, if muscle activation precedes arousal or is not accompanied by arousal during the end of the respiratory event, the muscle function is strong and the arousal threshold is high, which is suitable for surgery, otherwise, if arousal precedes muscle activation, the muscle function is weak or the arousal threshold is low, which is not suitable for surgery. Figure 1

[0022] Since the electromyography of the lower jaw can reflect muscle activity, and the electroencephalogram can reflect the activity of the cerebral cortex, it is used to evaluate arousal. Therefore, in the embodiments of the present application, the order of the changes of the electromyography of the lower jaw and the electroencephalogram during the end of the respiratory event is analyzed based on transfer entropy, and an arousal cause index that can quantify muscle function and arousal threshold at the same time is proposed, which is defined as: the proportion of the change of the electromyography of the lower jaw preceding the change of the electroencephalogram during the end of the OSA event of the patient. Based on the arousal cause index, a subtype suitable for surgery can be established, which helps to guide the clinical evaluation scheme and provides a new perspective for elucidating the physiological and pathological mechanisms of OSA.

[0023] In addition, in the embodiments of the present application, the arousal cause index is combined with clinical parameters of anatomy, PSG (polysomnography) parameters and CT parameters to construct an OSA surgical effect prediction model for predicting the surgical effect, which can directly predict the postoperative AHI (apnea-hypopnea index) of the patient before the operation according to the electromyography signal and the electroencephalogram signal of the OSA patient, provide objective parameters, and has important clinical value for judging whether the patient is suitable for receiving palatopharyngeal surgery.

[0024] ​It should be noted that the Changgeng sleep health data set is established in the embodiment of the application, and the data comes from the case data of the sleep center of Beijing Tsinghua Changgeng Hospital from 2020 to 2023.

[0025] Among them, the data inclusion criteria for OSA diagnosis and etiological typing are: (1) age 2-80 years old; (2) whole night PSG is performed. The exclusion criteria are: (1) previously diagnosed as respiratory failure; (2) severe obesity (BMI>40kg / m2); (3) sleep duration less than 4 hours.

[0026] Since the diagnostic criteria, etiology and treatment methods of adult OSA and pediatric OSA are different, the data set is divided into adult data set (19-80 years old) and pediatric data set (2-18 years old).

[0027] The data inclusion criteria for establishing the OSA surgical efficacy prediction model for surgical efficacy are: adult patients receiving palatopharyngeal surgery. Among them, the palatopharyngeal surgery includes uvulopalatopharyngoplasty (UPPP), uvuloplastic surgery and hard palate shortening. During the operation, all patients are routinely performed soft palate surgery, and then 70° endoscopic observation is performed through the nose to observe the degree of palatopharyngeal opening. If the palatopharyngeal airway is still collapsed or close to collapse, further hard palate shortening will be performed. The exclusion criteria are: (1) no follow-up data within six months; (2) previous OSA surgery history; (3) severe lung, nervous system, cardiovascular or mental disease.

[0028] The definition of surgical efficacy is: postoperative AHI is less than 20 times / hour, and the ratio of postoperative AHI to preoperative AHI is less than 50%.

[0029] First embodiment As Figure 2 shown, the first embodiment of the application provides a non-invasive obstructive sleep apnea arousal cause evaluation method, mainly including the following two parts: S10: calculating the arousal cause index of OSA patients; S20: predicting postoperative AHI of OSA patients.

[0030] Next, the above two parts will be described in detail: S10: calculating the arousal cause index of OSA patients.

[0031] It is known that neural signals have nonlinear and non-stationary characteristics, and linear characteristics are insufficient to fully explain the underlying neurophysiological activity mechanism in the process of complex brain function execution, and this problem can be improved by analyzing the nonlinear characteristics. The transfer entropy can be used to estimate the functional coupling strength and information transmission direction characteristics between the cortex and the muscle, and reveal the motor control and response mechanism between the cortex and the muscle. The greater the transfer entropy, the greater the coupling strength in the direction. The transfer entropy of X→Y is greater than that of Y→X, indicating that the information transmission of X to Y is stronger, and the change of X precedes that of Y in time.

[0032] The formula for calculating the transfer entropy is as follows: wherein, represents a joint probability distribution, which describes the probability of the simultaneous occurrence of Y t+1 , the history of Y and the history of X. The logarithmic term represents the proportion of the improvement in the conditional probability of predicting Y t+1 after adding the history information of X.

[0033] Based on the above calculation method of the transfer entropy, the awakening cause index of the OSA patient is calculated by the following steps in the embodiment of the present application: S11: Calculate the bidirectional transfer entropy of each OSA event.

[0034] The OSA event of the OSA patient is identified in advance (the specific identification method can be selected by itself, and is not specifically limited here). After the OSA event of the OSA patient is identified, on the one hand, the total number n of OSA events (determined by the PSG parameters of the patient) can be obtained; on the other hand, the end position of each OSA event can also be obtained.

[0035] Therefore, according to the calculation formula of the transfer entropy, the first transfer entropy from the myoelectric signal of the mandible to the electroencephalogram and the second transfer entropy from the electroencephalogram to the myoelectric signal of the mandible are calculated when the time difference is 1 second, by taking 5 seconds before and after the end position of each OSA event as the end period of the OSA event.

[0036] It can be understood that for the same OSA event, the first transfer entropy and the second transfer entropy jointly constitute the bidirectional transfer entropy of the OSA event.

[0037] S12: Determine the sequence of the myoelectric change and the electroencephalogram change in each OSA event.

[0038] Specifically, the first transfer entropy and the second transfer entropy are compared.

[0039] If the first transfer entropy is greater than the second transfer entropy, it indicates that the myoelectric change in the OSA event precedes the electroencephalogram change; otherwise, it indicates that the electroencephalogram change precedes the myoelectric change of the mandible.

[0040] S13: Count the number of OSA events in which the change in the mylohyoid muscle activity precedes the change in the EEG activity, denoted as m.

[0041] S14: Take m / n as the arousal cause index of the OSA patient.

[0042] In addition, in the embodiment of the present application, the correlation between the arousal cause index and the clinical parameters (age, BMI), AHI, apnea index (AI), hypopnea index (HI) and arousal threshold in OSA events is analyzed by calculating the Pearson correlation coefficient.

[0043] The results show that the arousal cause index is negatively correlated with BMI (body mass index) (the Pearson correlation coefficients in the adult data set and the child data set are -0.106, p<0.05; -0.096, p=0.31, respectively); and is negatively correlated with age, as shown in the following table, the index is significantly higher in children under 10 years old than in other age groups (0.91 [0.75, 1] vs. 0.82 [0.59, 1], 0.78 [0.63, 0.90], 0.80 [0.67, 0.90], 0.76 [0.67, 0.83] 0.73 [0.66, 0.89], all p<0.05). Figure 3A

[0044] In the adult data set, the arousal cause index is negatively correlated with AHI, AI and HI (the Pearson correlation coefficients are -0.151, -0.117 and -0.091, respectively, all p<0.05). The index is negatively correlated with the severity of OSA (Chi-square F=4.377, p=0.005), as shown in the following table. Figure 3B

[0045] To analyze the relationship between the arousal cause index and the arousal threshold, the drug-induced sleep endoscopy (DISE) data of 100 patients were collected, which included the mylohyoid muscle activity, the EEG and the esophageal pressure. The esophageal pressure can be used to calculate the arousal threshold. In the embodiment of the present application, the arousal threshold is the average of the lowest esophageal pressure before arousal. The esophageal pressure is usually negative, and the lower the esophageal pressure, the higher the corresponding arousal threshold.

[0046] The DISE was collected for an average of 18.9±9 minutes, and after excluding patients who had 2 or fewer OSA events during DISE, a total of 61 patients were included. The results are as follows. Figure 3C ​​As shown, the arousal etiology index was negatively correlated with the lowest esophageal pressure, with a Pearson correlation coefficient of -0.31 (p < 0.05). This result indicates that the arousal etiology index is positively correlated with the arousal threshold. When muscle function is strong, the arousal threshold is high; when arousal regulation is dominant, the arousal threshold is low.

[0047] Therefore, based on the correlation results that the arousal etiology index is negatively correlated with PSG parameters, BMI and age, and positively correlated with the arousal threshold, the rationality of the arousal etiology index proposed in the embodiments of the present invention is confirmed.

[0048] More preferably, in this embodiment of the invention, awakening etiology indicators and surgical effectiveness data corresponding to patients who have undergone surgery are obtained. Furthermore, based on the awakening etiology indicators and surgical effectiveness data of patients who have undergone surgery, an ROC curve predicting surgical efficacy based on the awakening etiology indicators is plotted. Finally, the area under the ROC curve is calculated, and the surgical subtype discrimination value is determined based on the maximum sum of sensitivity and specificity.

[0049] In this embodiment of the invention, based on the plotted ROC curve, the maximum sum of sensitivity and specificity was calculated to be 0.78. Therefore, using 0.78 as the surgical subtype discrimination value, the arousal etiology index of OSA patients can be compared with this surgical subtype discrimination value. If the arousal etiology index of an OSA patient is greater than the surgical subtype discrimination value, then the OSA patient is suitable for surgery; otherwise, surgery is not suitable.

[0050] In summary, by calculating the arousal etiology index of OSA patients, the muscle function of OSA patients can be accurately quantified. Furthermore, by combining the arousal etiology index with surgical effectiveness data, a surgical subtype differentiation value can be determined, thereby effectively screening patients suitable for surgical treatment and providing a new perspective for revealing the physiological and pathological mechanisms of OSA.

[0051] S20: Predict postoperative AHI in OSA patients.

[0052] After outputting the awareness etiology index for OSA patients in step S10, this awareness etiology index is combined with anatomically based clinical and CT parameters to form a training set. The training set is then input into various machine learning models for training. By comparing the predictive performance of these models, the machine learning model with the highest predictive performance is ultimately selected as the OSA surgical efficacy prediction model.

[0053] Below, in conjunction with Figure 4 Detailed explanation of the construction process of this OSA surgical efficacy prediction model: S21: Data Acquisition.

[0054] Specifically, based on historical patient surgery data, parameters in the surgery data with a correlation greater than a first threshold (for example: 0.3) with postoperative AHI are selected, and parameters with a correlation greater than a second threshold (for example: 0.8) between parameters are removed, as an anatomical input parameter. In an embodiment of the present application, the anatomical input parameter at least includes a clinical parameter, a PSG parameter and a CT parameter.

[0055] S22: Obtain the arousal cause index of the historical patient to cooperate with the anatomical input parameter to form a data set, and divide the data set into a training set and a test set.

[0056] S23: Input the training set into a plurality of machine learning models respectively for model training, and test the various machine learning models through the test set.

[0057] In an embodiment of the present application, the plurality of machine learning models at least include an artificial neural network (ANN) model, a support vector regression (SVR) model, a k-nearest neighbor (KNN) model, a random forest (RF) model, an XGBoost model and a linear regression model, but are not limited thereto.

[0058] S24: Evaluate the performance of the various machine learning models.

[0059] Specifically, the Pearson correlation coefficient, the root mean square error (RMSE) and the mean absolute error (MAE) are used to evaluate the regression performance of the various machine learning models after model testing; and the accuracy, sensitivity and specificity are used to evaluate the classification performance of the various machine learning models after model testing.

[0060] As shown in Table 1, the performance evaluation of the six machine learning models in the embodiment of the present application is as follows: Table 1 S25: Based on the regression performance and the classification performance of the various machine learning models, select the machine learning model with the highest comprehensive prediction performance as the OSA surgery efficacy prediction model.

[0061] From Table 1, it can be seen that the comprehensive prediction performance of the XGBoost model is optimal, the accuracy is 0.83±0.01, the sensitivity of surgical effectiveness is 0.86±0.02, and the specificity is 0.75±0.02. Therefore, in the embodiment of the application, the XGBoost model after model testing is preferably used as the OSA surgical efficacy prediction model.

[0062] Figure 5 The scatter plot of the postoperative AHI predicted value and the true value based on the anatomical factor and the arousal cause index model in the embodiment of the application is shown. It can be seen from the scatter plot that in the XGBoost model, the difference between the postoperative AHI predicted value and the true value is distributed around 0, and is normally distributed. Figure 6 The difference histogram of the postoperative AHI predicted value and the true value based on the anatomical factor and the arousal cause index model is shown.

[0063] In order to study the importance of clinical characteristics, PSG parameters, CT parameters and arousal cause indexes, the SHAP value of each feature in the XGBoost model is calculated in the embodiment of the application. The larger the SHAP value, the more important the feature. The results are shown in Figure 7 It can be seen from the scatter plot that preoperative AHI and BMI are the most important prediction factors, followed by arousal cause indexes and airway length. The important prediction effect of the arousal cause index is verified by the results.

[0064] It can be understood that after the construction of the OSA surgical efficacy prediction model is completed, the clinical parameters, CT parameters, PSG parameters and arousal cause indexes are input into the OSA surgical efficacy prediction model, so as to output the postoperative AHI prediction result of the OSA patient.

[0065] As shown in Figure 8 On the basis of the above embodiment, the embodiment of the application further provides an OSA precise evaluation software integrating OSA diagnosis, respiratory event identification, arousal cause index calculation and surgical efficacy prediction. The architecture of the software adopts a layer architecture, mainly including a user interface, a business layer, a database and a running environment. The database saves the data input at present and in the future.

[0066] Python language is used as a development tool, a standard GUI toolkit Tkinter is used to realize user interface design, and a matplotlib visualization library is used to realize data visualization function. Two running environment configurations are as follows: (1) Operating system: Windows 64-bit, CPU processor: Intel(R) Core (TM) i7-5500U CPU @2.40GHz, memory: 8G (2) Operating system: GNU / Linux 5.15.0-94 64-bit system, CPU processor: Intel(R) Xeon (R) CPU E5-2620 v4 @ 2.10GHz, memory: 64G Figure 9 The user interface is for OSA diagnosis and respiratory event recognition. The operation instructions are as follows: (1) After opening the software, load the mylohyoid muscle electrical signal and select the patient number. (2) Click the

Show Muscle Electrical

Start Recognition

Clear

[0067] (4) Click the

Calculate Awakening Cause Index

[0068] (5) Click the

Palatopharyngeal Surgery Efficacy Prediction

Predict

[0069] When the surgery efficacy prediction is completed, click the

Clear

[0070] Second embodiment As shown in Figure 11 , on the basis of the above-mentioned first embodiment, the second embodiment of the present application provides a non-invasive obstructive sleep apnea awakening cause evaluation system, which comprises a data acquisition unit 1, an electromyographic recognition unit 2, a calculation unit 3, and an output unit 4. Among them, The data acquisition unit 1 is used to acquire the mylohyoid muscle electrical signal, the electroencephalogram signal, the clinical parameter based on anatomy, the PSG parameter, and the CT parameter of the OSA patient.

[0071] The electromyographic recognition unit 2 is preset with the above-mentioned OSA recognition model and is connected with the data acquisition unit 1, so as to recognize the mylohyoid muscle electrical signal, thereby obtaining the end position of each OSA event of the OSA patient.

[0072] The computing unit 3 is connected with the myoelectric identification unit 2 and the data acquisition unit 1, to calculate the arousal cause index of the OSA patient based on the ending position of each OSA event and the electroencephalogram of the OSA patient.

[0073] The output unit 4 is preset with the OSA surgical efficacy prediction model and is connected with the computing unit 3 and the data acquisition unit 1, to input the clinical parameters, CT parameters, PSG parameters and arousal cause index of the patient into the OSA surgical efficacy prediction model, so as to output the postoperative AHI prediction result of the OSA patient.

[0074] It can be understood that the above-mentioned functional units are only one specific implementation form of each step of the non-invasive obstructive sleep apnea arousal cause evaluation method in the above-mentioned embodiments, but are not limited to the functions and connection modes of the functional units, and in other embodiments, the functions and connection modes of the functional units can be adaptively adjusted to realize each step of the non-invasive obstructive sleep apnea arousal cause evaluation method.

[0075] Third embodiment As Figure 12 shown, on the basis of the non-invasive obstructive sleep apnea arousal cause evaluation method, the third embodiment of the present application further provides a non-invasive obstructive sleep apnea arousal cause evaluation system. The system comprises one or more processors and memories. Wherein, the memory is coupled with the processor, for storing one or more programs, when the programs are executed by the processor, so that the processor realizes the non-invasive obstructive sleep apnea arousal cause evaluation method in the above-mentioned embodiments.

[0076] Wherein, the processor is used to control the overall operation of the system, to complete all or part of the steps of the above-mentioned non-invasive obstructive sleep apnea arousal cause evaluation method. The processor can be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable logic gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processing (DSP) chip, etc. The memory is used to store various types of data to support the operation of the system, which can include, for example, instructions for operating any application or method on the system, and application-related data. The memory can be realized by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, etc.

[0077] In one example embodiment, the system can be embodied by a computer chip or entity, or by a product with certain functions, for implementing the non-invasive obstructive sleep apnea arousal cause assessment method described above, and achieving the technical effects consistent with the method described above. A typical embodiment is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0078] In another example embodiment, the present application also provides a computer readable storage medium comprising program instructions, which, when executed by a processor, implement the steps of the non-invasive obstructive sleep apnea arousal cause assessment method in any one of the embodiments described above. For example, the computer readable storage medium can be the memory described above comprising program instructions, which can be executed by the processor of the non-invasive obstructive sleep apnea arousal cause assessment system to complete the non-invasive obstructive sleep apnea arousal cause assessment method described above, and achieve the technical effects consistent with the method described above.

[0079] It should be noted that the above embodiments are only illustrative. The technical solutions of each embodiment can be combined, and all are within the protection scope of the present application.

[0080] It should be understood that the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0081] The non-invasive obstructive sleep apnea arousal cause assessment method and system provided by the present application are described in detail above. Any obvious modification made by a person of ordinary skill in the art without departing from the essential content of the present application will constitute an infringement of the right of the present application, and the person will bear the corresponding legal responsibility.

Claims

1. A method of non-invasive obstructive sleep apnoea-hypopnoea event assessment, characterised in that The method comprises the following steps: obtaining the end position of each OSA event of an OSA patient; obtaining the electromyography signal and the electroencephalogram signal of the OSA patient, and calculating the proportion of the change in the electromyography signal preceding the change in the electroencephalogram signal during the end period of all OSA events based on the end position of each OSA event as the arousal cause index of the OSA patient; obtaining the clinical parameters, PSG parameters and CT parameters of the OSA patient, wherein the clinical parameters at least include the body mass index, the PSG parameters at least include the apnea-hypopnea index, the arousal index and the time of blood oxygen below 90%, and the CT parameters at least include the vertical distance between the lower edge of the mandible and the lower edge of the hyoid bone and the airway length; inputting the clinical parameters, PSG parameters, CT parameters and the arousal cause index into a preset OSA surgery effect prediction model to output the postoperative apnea-hypopnea index prediction result of the OSA patient; wherein a data set is formed based on the clinical parameters, CT parameters, PSG parameters and arousal cause index of a plurality of historical patients, and the data set is input into a plurality of machine learning models for model training, and the machine learning model with the highest prediction performance is selected as the OSA surgery effect prediction model by comparing the prediction performance of the plurality of machine learning models.

2. The non-invasive obstructive sleep apnea-hypopnea event etiology assessment method of claim 1, wherein The arousal cause index is obtained by the following steps: taking the 5-second time before and after the end position of each OSA event as the end period of the OSA event, calculating the first transfer entropy from the electromyography signal to the electroencephalogram signal and the second transfer entropy from the electroencephalogram signal to the electromyography signal when the time difference is 1 second; if the first transfer entropy is greater than the second transfer entropy, it indicates that the change in the electromyography signal precedes the change in the electroencephalogram signal in the OSA event; otherwise, it indicates that the change in the electroencephalogram signal precedes the change in the electromyography signal; counting the number of OSA events in which the change in the electromyography signal precedes the change in the electroencephalogram signal in all OSA events, denoted as m; taking m / n as the arousal cause index of the OSA patient, wherein n is the total number of OSA events identified based on the electromyography signal of the OSA patient.

3. The non-invasive obstructive sleep apnea-hypopnea event etiology assessment method of claim 2, wherein Further comprising: obtaining the corresponding arousal cause index and surgery effectiveness data of the historical patients who have undergone surgery; based on the arousal cause index and surgery effectiveness data of the historical patients who have undergone surgery, drawing the ROC curve for predicting the surgery effect based on the arousal cause index with 1-specificity as the horizontal coordinate and sensitivity as the vertical coordinate; calculating the area under the ROC curve, and determining the surgery subtype discrimination value according to the maximum sum of sensitivity and specificity; comparing the arousal cause index of the OSA patient with the surgery subtype discrimination value, if the arousal cause index of the OSA patient is greater than the surgery subtype discrimination value, the OSA patient is suitable for surgery; otherwise, the OSA patient is not suitable for surgery.

4. The non-invasive obstructive sleep apnea arousal cause evaluation method of claim 3, wherein: the maximum sum of sensitivity and specificity under the ROC curve is 0.

78.

5. The non-invasive obstructive sleep apnea-hypopnea event etiology assessment method of claim 1, wherein The OSA surgical efficacy prediction model was constructed through the following steps: Based on historical patient surgical data, parameters in the surgical data with a correlation greater than a first threshold to the postoperative apnea-hypopnea index are selected, and parameters with a correlation greater than a second threshold are removed, and these are used as anatomical input parameters; wherein, the anatomical input parameters include at least clinical parameters, PSG parameters, and CT parameters; Historical patients' arousal etiology indicators are obtained and used in conjunction with the anatomical input parameters to form a dataset, which is then divided into a training set and a test set. The training set is input into various machine learning models for model training, and the test set is used to test each machine learning model. The regression performance of each machine learning model after model testing was evaluated using Pearson correlation coefficient, root mean square error, and mean absolute error, respectively, and the classification performance of each machine learning model after model testing was evaluated using accuracy, sensitivity, and specificity, respectively. Based on the regression and classification performance of each machine learning model, the machine learning model with the highest overall predictive performance is selected as the OSA surgery efficacy prediction model.

6. The non-invasive method for assessing the etiology of obstructive sleep apnea awakening as described in claim 5, characterized in that: The machine learning model includes at least one or more of the following: artificial neural network model, support vector regression model, k-nearest neighbor model, random forest model, XGBoost model, and linear regression model.

7. The non-invasive method for assessing the etiology of obstructive sleep apnea awakening as described in claim 1, characterized in that: The arousal etiology index is negatively correlated with the PSG parameter, the body mass index and age, and positively correlated with the arousal threshold.

8. A non-invasive obstructive sleep apnea-hypopnea event cause assessment system for implementing the non-invasive obstructive sleep apnea-hypopnea event cause assessment method according to any one of claims 1 to 7, characterized by include: The data acquisition unit is used to acquire mandibular electromyography (EMG) signals, electroencephalogram (EEG) signals, anatomical-based clinical parameters, PSG parameters, and CT parameters of OSA patients. An electromyography (EMG) recognition unit is pre-configured with an OSA recognition model and connected to the data acquisition unit to recognize the mandibular EMG signals, thereby obtaining the end position of each OSA event in the OSA patient. The calculation unit, connected to the electromyography recognition unit and the data acquisition unit, calculates the proportion of mandibular electromyography changes preceding brainwave changes during the end of all OSA events based on the end position of each OSA event and the EEG signal of the OSA patient, as an indicator of the cause of arousal in the OSA patient. The output unit is pre-set with an OSA surgical efficacy prediction model and is connected to the calculation unit and the data acquisition unit. It is used to input the clinical parameters, CT parameters, PSG parameters and the awakening etiology indicators into the OSA surgical efficacy prediction model, thereby outputting the postoperative AHI prediction results of the OSA patient. Specifically, a dataset is formed based on clinical parameters, CT parameters, PSG parameters, and arousal etiology indicators of multiple historical patients. The dataset is then input into various machine learning models for model training. By comparing the predictive performance of the various machine learning models, the machine learning model with the highest predictive performance is selected as the OSA surgery efficacy prediction model.

9. The evaluation system as described in claim 8, characterized in that: The OSA recognition model is formed by jointly training an LSTM model and a CRF model.

10. A non-invasive obstructive sleep apnea-hypopnea event etiology assessment system characterized by It includes a processor and a memory, wherein the processor reads a computer program from the memory for implementing the non-invasive obstructive sleep apnea awakening etiology assessment method according to any one of claims 1 to 7.

Images