Osahs detection method, system and device, and medium

Abstract

Blood oxygen data are obtained and preprocessed, and analysed using a one-dimensional neural network (NN) classifier to obtain first-stage classification into normal respiration, suspected hypopnea, o

Description

JtiVyJnLli 1CAL Jr 1LLLJ The present invention relates to the technical field of data detection, and in particular, to an 5 OSAHS detection system and device, and a medium. BACKGROUND An obstructive sleep apnea hypopnea syndrome (OSAHS) is currently the most common sleep-disordered breathing, and is characterized by repetitive episodes of airflow reduction (hypopnea) or 10 cessation due to partial or complete upper airway collapse during sleep, resulting in frequent desaturation, hypercapnia, and sleep fragmentation. With the development of obesity and aging, the prevalence rate of the OSAHS has increased rapidly. It is estimated that 936 million people aged SO-69 worldwide suffer from this disease. The OSAHS can cause acute respiratory failure or sudden death, increasing the prevalence of hypertension, stroke, coronary heart disease, senile dementia, and diabetes, and increasing the mortality of patients with chronic obstructive pulmonary disease, renal insufficiency, and heart failure. It has become a global health problem. Polysomnography (PSG), first reported in 1974, is still the gold standard for diagnosing the OSAHS, that is, the polysomnography is used in a sleep monitoring room to continuously and synchronously collect and record multiple sleep-related physiological parameters, including an 20 electroencephalogram, an electro-oculogram, an electromyogram, an electrocardiogram, an oronasal airflow, a pulse oxygen saturation (SpO2), thoracoabdominal breathing movements, a body position, snoring, etc.; according to the latest edition of American Academy of Sleep Medicine (AASM) Manual for the Scoring of Sleep and Associated Events: Rules, Terminologies, and Technical Specifications, sleep experts interpret and analyze these parameters, and finally obtain the apnea-hypopnea index 25 (AHI) for describing the frequency of a sleep-related respiratory event, and use it as an indicator for diagnosing the OSAHS and classifying the severity of this disease[7]: AHI<5 indicates non-OSAHS, 5<AHI<15 indicates mild, 15<AHI<30 indicates moderate, and AHI>30 indicates severe. However, the above-mentioned whole process has the disadvantages of high technical threshold, heavy patient burden, high price, time and labor consumption, and being prone to errors, which significantly affects 30 the practicability and availability of OSAHS diagnosis resources. According to the AASM recommended rules, an apnea event is defined as a decrease >90% in 21 05 24 respiratory airflow compared with a baseline value, which lasts for at least 10s; a hypopnea event is defined as a decrease >30% in respiratory airflow compared with the baseline value, which lasts for at least 10s, accompanied with oxygen saturation reduction >3% or awakening; AHI is defined as the sum of the number of apnea and hypopnea events in the total sleep time per hour, and is an important 5 indicator for conveying information about sleep-disordered breathing. It is worth mentioning that the initial research on respiration events only focused on the apnea, but it is soon found that an airflow decrease is as serious as the damage to the body and the apnea, and has clinical significance. On the one hand, the identification of the hypopnea event is related to accurate calculation of an AHI value, directly affecting the diagnosis and treatment decisions of the OSAHS. On the other hand, in some 10 OSAHS patients, a main respiration event recorded by a PSG is not apnea but hypopnea, and recognizing a hypopnea event is beneficial to individualized diagnosis and treatment. Studies have demonstrated that extremely obese subjects (BMI>45kg / m2) with the OSAHS show very frequent hypopneas and relatively few apneas, which proves that the hypopneas and apneas may have different mechanisms, and may lead to specific treatment for hypopnea-predominant sleep-disordered 15 breathing. However, most studies at present only identify the apnea event and ignore the hypopnea event, which cannot fully and accurately describe sleep-disordered breathing, likely leading to underestimation or even missed diagnosis. In view of this, how to achieve accurate and low-cost detection of OSAHS is an urgent problem to be solved. SUMMARY In view of this, embodiments of the present invention provide an OSAHS detection system and device, and a medium, which can achieve accurate and low-cost OSAHS detection. 25 In an aspect, the embodiments of the present invention provide an OSAHS detection system, comprising: a first module for obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice; a second module for analyzing the blood oxygen signal data slice by using a one-dimensional 30 neural network classifier to obtain first-stage classification results; wherein the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; and a third module for analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discrimination criteria; the target fragment comprises fragments corresponding to the blood oxygen 5 signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage classification results include normal respiration, hypopnea and apnea. Optionally, obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice comprises: obtaining original blood oxygen signal data of a target object in a preset time period; 10 screening, based on a preset proportion threshold of a blood oxygen artifact, the original blood oxygen signal data to obtain target blood oxygen signal data; slicing the target blood oxygen signal data based on a preset duration; and down-sampling and normalizing the target blood oxygen signal data obtained after the slicing, to obtain blood oxygen signal data slice. Optionally, the OSAHS detection system is for: LO creating one-dimensional neural network classifier based on the blood oxygen signal data of the C* ^ labelled sleep-related respiratory event. Optionally, creating one-dimensional neural network classifier based on blood oxygen signal data of the labelled sleep-related respiratory event comprises: 20 determining training samples based on the blood oxygen signal data of the labelled sleep-related respiratory event; providing a neural network model with two branches based on a one-dimensional convolution operator, a one-dimensional maximum pooling operator, an activation function and a fully connected layer, performing classification training on the neural network model based on the training samples, 25 and adjusting the neural network model based on training results to obtain a one-dimensional neural network classifier, wherein the two branches of the neural network model each comprise several onedimensional convolution operators, a one-dimensional maximum pooling operator and an activation function, and the two branches of the neural network model are connected in the fully connected layer. 30 Optionally, the one-dimensional neural network classifier comprises two branches, and analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results comprises: 21 05 24 performing fragment extension according to each fragment in the blood oxygen signal data slice to obtain a first branch input; and performing fragment combination to obtain a second branch input; performing convolution operation, activation response and dimensionality reduction respectively on the first branch input and the second branch input to correspondingly obtain a first 5 feature map and a second feature map; splicing the first feature map and the second feature map according to a channel dimension to obtain a third feature map; and performing fully connected classification on the third feature map to obtain the first-stage classification results. 10 Optionally, the OSAHS detection system is for: setting discrimination rules for the scores of the blood oxygen signal statistical indexes based on blood oxygen saturations, blood oxygen signal variances and blood oxygen minimum values; and creating a classifier based on rule matching according to the discrimination rules. Optionally, analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching comprises: performing first scoring processing according to a reduction amplitude of the blood oxygen saturation of the blood oxygen signal data slice of the target fragment; performing second scoring processing according to a fluctuation range of the blood oxygen signal variance of the blood oxygen signal data slice of the target fragment; 20 performing third scoring processing according to a reduction amplitude of the blood oxygen minimum value of the blood oxygen signal data slice of the target fragment; determining the scores of the blood oxygen signal statistical indexes by means of the first scoring processing, the second scoring processing and the third scoring processing; and obtaining second-stage classification results according to the scores of the blood oxygen signal 25 statistical indexes. In yet another aspect, the embodiments of the present invention provide an OSAHS detection device, comprising a processor and a memory; the memory being configured to store a program; and the processor executing the program to implement a method comprising: 30 obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice; analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results; wherein the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; 5 and analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discrimination criteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice 10 with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage classification results include normal respiration, hypopnea and apnea. In yet another aspect, the embodiments of the present invention provide a computer-readable storage medium, wherein the computer-readable storage medium stores a program, and the program is executed by a processor to implement a method comprising:- obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal C\l data to obtain blood oxygen signal data slice; LO analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results; wherein the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory 20 event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; and analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discrimination 25 criteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage classification results include normal respiration, hypopnea and apnea. The embodiments of the present invention further disclose a computer program product or a computer program, wherein the computer program product or the computer program comprises 30 computer instructions that are stored in a computer-readable storage medium. A processor of a computer device may read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions such that the computer device executes the foregoing method. 21 05 24 In the embodiments of the present invention, the OSAHS detection method comprises: obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice; analyzing the blood oxygen signal data slice by using a onedimensional neural network classifier to obtain first-stage classification results; wherein the one-5 dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; and analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen 10 signal statistical indexes as discrimination criteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage classification results include normal respiration, hypopnea and apnea. The detection data of the present invention is only based on blood oxygen signals, so that signal acquisition can be simplified, and the detection costs can be effectively 15 reduced. Meanwhile, by means of the one-dimensional neural network classifier and the classifier based on rule matching, the first-stage classification is first achieved with the neural network based on learning, the generalization of the classification results is ensured, and the second-stage classification is then achieved in combination with the scores of the blood oxygen signal statistical indexes, classification refinement and deviation correction are achieved, and accurate classification results are obtained. The present invention can achieve accurate and low-cost OSAHS detection. BRIEF DESCRIPTION OF DRAWINGS In order to explain the technical solutions in the examples of the present invention more clearly, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, 25 the drawings in the following description are only some of the examples of the present invention, and for those of ordinary skill in the art, other drawings would also have been obtained from these drawings without any inventive effort. Fig. 1 is a schematic flowchart of an OSAHS detection method according to the embodiments of the present invention; 30 Fig. 2 is a schematic overall flowchart of an OSAHS detection method according to the embodiments of the present invention; Fig. 3 is a schematic structural diagram of a one-dimensional neural network classifier according to the embodiments of the present invention; Fig. 4 is a schematic diagram of blood oxygen variances in different states according to the embodiments of the present invention; and Fig. 5 is a schematic diagram of blood oxygen minimum values in different states according to the embodiments of the present invention. DETAILED DESCRIPTION In order to make the object, the technical solution and the advantages of the present invention clearer, the present invention will be further illustrated in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that specific embodiments 10 described herein are only intended to explain the present invention, and are not taken to limit the present invention. First, it needs to be explained that some scholars have currently classified hypopnea and apnea events in one category by means of OSAH detectors based on blood oxygen, so as to reduce the complexity of algorithm implementation and improve the accuracy rate. However, due to confusion of the apnea and the hypopnea. it is in.poss.ble to distinguish whether a eertain patients oxygen drop is caused by the hypopnea or the apnea. LO However, some scholars try to distinguish the apnea from the hypopnea with different machine learning methodologies. In the prior art, an OSAH is detected with a sparse representation structured C\j dictionary learning method, but the effect is not good. In the Sleep Heart Health Study database 20 (SHHS), the hypopnea sensitivity is only 22.78%. Some scholars further participate in the detection with the gold standard of OSAH diagnosis, that is, use airflow signals to assist in hypopnea identification. In the prior art, there is an OSAH detector designed using a two-dimensional convolutional neural network (2DCNN) combined with a fully connected neural network (FCN), which detects the OSAH in children in combination with 25 airflow and blood oxygen signals. In the Childhood Adenotonsillectomy Trial database (CHAT), good results are achieved. However, it is worth noting that the acquisition cost of the airflow signals is higher than that of the blood oxygen signals by the cost difference of about ten times. Different from the above-mentioned method, the present invention provides a second-stage classification model based on a single-channel blood oxygen signal, which achieves automatic and 30 accurate detection on two respiration events of apnea and hypopnea: (1) a model input is only a subject’s blood Oxygen signal, taking into account the simplicity of signal acquisition and patient comfort; (2) a second-stage model integrates statistical information of a neural network and a blood oxygen time domain, can improve the identification rate of an algorithm for hypopnea, is low in 21 05 24 training complexity, and can complete the reasoning process of low-power-consumption equipment such as smart bracelets. In one aspect, with reference to Fig. 1, the embodiments of the present invention provide an OSAHS detection system for performing the followings steps. 5 At S100, original blood oxygen signal data is obtained and subjected to data preprocessing to obtain blood oxygen signal data slice. It should be noted that in some embodiments, the original blood oxygen signal data of a target object in a preset time period is obtained; based on a preset proportion threshold of a blood oxygen artifact, the original blood oxygen signal data is screened to obtain target blood oxygen signal data; 10 the target blood oxygen signal data is subjected to data slicing based on a preset duration; target blood oxygen signal data obtained after the slicing is down-sampled and normalized to obtain blood oxygen signal data slice. Specifically, first, in order to ensure that SpCh signals (that is, blood oxygen signals) are collected in a sleeping state of a subject, the embodiments of the present invention use the data obtained from 0:00 midnight to 6:00 am, which is a preferable time period. Second, in order to reduce the interference of noise signals, only the data whose blood oxygen artifact accounts for less than 1 % of the total length of a record is selected, and the remaining blood oxygen artifact is eliminated by means of nearest neighbor interpolation. Third, the data is sliced according to a length of 1 minute. Finally, the sampling frequency of the SpO2 signals in a database is 10 Hz, and the blood oxygen time 20 series is down-sampled from 10 Hz to 1 Hz with an open-source algorithm, that is, a largest triangle three buckets (LTTB) down-sampling algorithm, and Z-Score normalization is performed, that is, a mean value of all SpO? data values is subtracted from the data values and the result is divided by a variance. At S200, the blood oxygen signal data slice is analyzed by using a one-dimensional neural 25 network classifier to obtain first-stage classification results. It should be noted that the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event. The first-stage classification results include normal respiration, suspected hypopnea, and apnea; In some embodiments, the system is for implementing the following steps. A one-dimensional 30 neural network classifier is created based on the blood oxygen signal data of the labelled sleep-related respiratory event, wherein training samples are determined based on the blood oxygen signal data of the labelled sleep-related respiratory event. A neural network model with two branches is provided based on a one-dimensional convolution operator, a one-dimensional maximum pooling operator, an activation function and a fully connected layer, is subjected to classification training based on the training samples, and is adjusted based on training results to obtain a one-dimensional neural network classifier; wherein the two branches of the neural network model each comprise several onedimensional convolution operators, a one-dimensional maximum pooling operator and an activation 5 function. The two branches of the neural network model are connected in the fully connected layer. Specifically, as shown in Fig. 2, the database used in the present invention is from the OSAHS clinical scientific research integrated big data platform of the First Affiliated Hospital of Sun Yat-sen University. Algorithms are trained, validated and tested using only SpO? single-channel data. The information of a total of 450 subjects are used. After data cleaning, the remaining 358 pieces of data 10 are available and are divided into a training set (i.e., training samples), a validation set and a test set according to the ratio of 6:2:2, that is to say, the training set contains 218 pieces of data, the validation set contains 70 pieces of data, and the test set contains 70 pieces of data. Due to poor machine connection quality and sweating or turning over of the patients at night, it is possible to cause a sensor to be in poor contact or even drop, resulting in SpO2 signal interruption. This case is called the artifact. When the length of the artifact exceeds 10% of the length of all-night signals, this record is discarded; when the length of the artifact is less than 10%, the artifact is erased by means of nearest neighbor linear interpolation, and the whole process is called data cleaning. Similarly, the training set also undergoes data preprocessing as described at step SI00. At the third step of the data preprocessing, each fragment having a 1-minute length after data slicing is labelled with a respiration event, and these fragments are labelled as normal (N), apnea (A) and hypopnea (H). Then, the neural network model is trained and adjusted based on the preprocessed training set to finally obtain the onedimensional neural network classifier. An input of the neural network model has two branches, and the input signals are blood oxygen signals down-sampled to 1 Hz. A detailed structure is shown in Fig. 3. In some specific embodiments, the tail ends of the two branches use Flatten as a feature map 25 flattening method. In some embodiments, analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results comprises the following steps. According to each fragment in the blood oxygen signal data slice, fragment extension is performed to obtain a first branch input. Fragment combination is performed to obtain a second branch input. 30 The first branch input and the second branch input are each subjected to convolution operation, activation response and dimensionality reduction respectively to correspondingly obtain a first feature map and a second feature map. The first feature map and the second feature map are spliced according to a channel dimension to obtain a third feature map. The third feature map is subjected to fully connected classification to obtain the first-stage classification results. 21 05 24 Specifically, as shown in Fig. 3: (1) The first input branch is denoted as InputlSp02. Typically, during sleeping, a blood oxygen saturation shows a downward trend with a time increase of the respiration event, but the apnea and the hypopnea do not always occur simultaneously each time. Considering that an oxygen desaturation 5 (OD) has a delay of 1-10 seconds compared with the respiration event, with the delay time varying from person to person, in order to offset the delay, an input length is extended from the standard 1 minute to 1 minute and 20 seconds (i.e., fragment extension), totaling 80 seconds. The extended 20 seconds of the current minute are the first 20 seconds of the next minute, that is, there is a signal window overlap of 20 seconds, so that a blood oxygen drop signal caused by a sleep-related 10 respiratory event occurring after a current minute detection window can also be captured. The blood oxygen signal fragments then undergo three one-dimensional convolution operators (ConvlD), which is intended to extract useful features from the input, the calculation for feature extraction is omitted by sharing weights and perceptual fields by means of convolution operations. A one-dimensional feature map extracted by the one-dimensional convolution operator is denoted as F-^ and the weight of the operator is denoted as ZSCORE normalization generate negative values. In order to ensure the responsiveness of the neural network to the negative values, LeakyReLU is used as an activation function herein, which is also responsive to negative values, different from ReLU which does not respond to negative values at all. In addition, dimensionality reduction is performed using the maximum pooling (MaxpoolingID), further reducing the quantity of parameters of the neural network, so as to reduce the overhead of forward reasoning. A data expression of the first branch is as follows: Fi =^^^^02,^). (2) The second input branch is denoted as Input2SpO2. For patients with severe OSAHS, their sleep-related respiratory events occur frequently and are often concentrated. For example, if a sleep- 25 related respiratory event occurs within the current 1 minute, it is very likely to occur again in the next 1 minute, that is, the occurrence of the sleep-related respiratory events is quasi-periodical. In order to capture this quasi-periodicity and improve the accuracy of prediction, the second input branch is designed to read blood oxygen signals in a larger range of the current 1 minute and 2 minutes respectively before and after the current 1 minute, totaling 300 seconds (that is, the combination of 30 fragments). Similar to the first input branch, the second input branch also undergoes a onedimensional convolutional neural network. A one-dimensional feature map extracted by the onedimensional convolution operator is denoted as F2, and a weight of the operator is denoted as d2. A data expression of the second branch is as follows: 21 05 24 F2 =f(lnput2SpO2,e2). Later, the feature maps Ft and F2 extracted by the two branches are spliced according to the channel dimension, and introduced to a fully connected classification networkDapnea (including a fully connected layer Dense) to complete first-stage classification, and 03 represents a weight of the 5 fully connected classification network. An output of the one-dimensional neural network classifier is normal and suspected hypopnea (N&H) or apnea (A), the N&H is classified in the second stage to distinguish a normal respiration fragment from a suspected hypopnea fragment. Data expressions are as follows: Fconcat =fVA f2), ^apnea ~^^'concat> ^3)- At S300, the blood oxygen signal data slice of a target fragment is analyzed by using a classifier based on rule matching, to obtain second-stage classification results. It should be noted that the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discrimination criteria. The target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea. The second-stage classification results include normal respiration, hypopnea and apnea. In some embodiments, the system is for implementing the following steps. Discrimination rules for the scores of the blood oxygen signal statistical indexes are set based on blood oxygen saturations, 20 blood oxygen signal variances and blood oxygen minimum values. Based on the discrimination rules, a classifier based on rule matching is created. Analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching comprises the following steps. First scoring processing is performed according to a reduction amplitude of the blood oxygen saturation of the blood oxygen signal data slice of the target 25 fragment. Second scoring processing is performed according to a fluctuation range of the blood oxygen signal variance of the blood oxygen signal data slice of the target fragment. Third scoring processing is performed according to a reduction amplitude of the blood oxygen minimum value of the blood oxygen signal data slice of the target fragment. The scores of the blood oxygen signal statistical indexes are determined by means of the first scoring processing, the second scoring 30 processing and the third scoring processing. Second-stage classification results are obtained according to the scores of the blood oxygen signal statistical indexes. Specifically, peak airflow curves of some hypopnea events have a smaller reduction amplitude than a baseline value, and the blood oxygen drop caused is not obvious, and characteristics thereof are difficult to be captured by the neural network. Therefore, the embodiments of the present invention introduce the second-stage classifier based on rule matching configured to identify a hypopnea respiration event. This classifier uses the scores of the blood oxygen signal statistical indexes as the discrimination criteria: a score is provided to the blood oxygen signal fragment meeting the statistical 5 indexes, and when the score reaches a certain level, the fragment is judged as a hypopnea event. The three scoring criteria will be described below. (1) 3% OD. If the drop of the blood oxygen saturation is greater than or equal to 3% compared with the baseline value, 1 point is scored, otherwise, no point is scored (i.e., the first scoring processing), which is also one of the rules for interpreting hypopnea recommended in the AASM 10 Manual for the Scoring of Sleep and Associated Events: Rules, Terminologies, and Technical Specifications. (2) Blood oxygen signal variances. Variance is a good measure of signal fluctuation. Normal people, whether they are awake or asleep, have very little blood oxygen fluctuations. However, when a sleep-related respiratory event occurs, due to airway collapse leading to reduced inspiratory airflow, ^5 the OSAHS patients cannot inhale sufficient oxygen, causing immediate blood oxygen drop and C\j normal value deviation, thereby increasing the variances. Fig. 4 shows the blood oxygen variances in LO different states. When the variance is greater than 0.5, 1 point is scored, which is also an upper quartile C* ^ line in a normal respiration state and a lower quartile line in a hypopnea state; when the variance is greater than 9, 2 points are scored, which is also an upper limit of the blood oxygen variance during OJ hypopnea, indicating that a respiration event more serious than hypopnea may have occurred at this case, that is, apnea; when the variance is less than 0.5, no point is scored. The above description shows the details of the second scoring processing. (3) Minimum values of blood oxygen. Fig. 5 shows minimum values of blood oxygen in different states. For normal people, the blood oxygen less than 94% is regarded as hypoxemia, but 25 the embodiments of the present invention aim to distinguish a normal fragment from a hypopnea fragment as much as possible. Therefore, when the minimum blood oxygen value is less than 90%, 1 point is scored, otherwise, no point is scored (that is, the third scoring processing). When the total score is greater than or equal to 2, it indicates that the blood oxygen fragment is very likely a hypopnea event; when the total score reaches 4, it indicates that the blood oxygen 30 fragment is likely an apnea event, but it is not labelled by the first-stage neural network, the fragment needs to be labelled again as apnea, otherwise, it is regarded as normal respiration. The OSAH detection and evaluation indexes mainly include accuracy (Acc), sensitivity (Sens), specificity (Spec), the number of positive samples in samples predicted to be true (Prec) andFi value, which are calculated according to the following expressions: 21 05 24 wherein TP, TN, FP and FNrepresent "true positive", "true negative", "false positive" and "false negative" respectively. In multiple classification tasks (referring to the number of classifications 5 greater than 3), the embodiments of the present invention measure each index in the form of macroaverage, that is, the indexes of each category are calculated, and their mean value is then calculated. In some specific embodiments of the present invention, experiments are performed to validate the effectiveness of the above method. Specifically, the provided algorithm is tested with a test set in the OSAH database provided by the First Affiliated Hospital of Sun Yat-sen University. We ensure that the test data does not undergo training and verification stages of the algorithm, and is only used for testing the second-stage model. Our test results are shown in Table 1, and the accuracy may reach 80.28%. Table 1 Accuracy (%) Sensitivity (%) Specificity (%) Fl score 80.28 75.41 90.84 0.8028 It is true that the accuracy of detecting OSAHS using airflow signals has reached more than 90% 15 at present, but the acquisition of airflow signals is not only expensive, but also requires thermal sensors and disposable pressure airflow tubes, which seriously affect the sleep quality of patients. The thermal sensors and the pressure airflow tubes must be placed at the mouths and noses of the patients and inserted into the nasal cavities, bringing extremely poor experience. The SpO2 signals used in the embodiments of the present invention may be obtained from a medical-grade finger pulse 20 oxygen sensor, or from cheap wearable devices such as smart bracelets, and only require an ultra-low signal resolution of I Hz. Low-resolution signals (sampling rate equal to 1 Hz) may be used directly, and high-resolution signals (sampling rate greater than 1 Hz) may be down-sampled for use. Therefore, the present invention provides a cheaper, more flexible and comfortable detection method while maintaining high detection accuracy. In summary, the present invention uses two-stage classification detection. At the first stage, the one-dimensional neural network classifier is used for preliminary classification to distinguish the apnea from other labeled data. At the second stage, based on rule matching, statistical information of 5 the blood oxygen time series is used for distinguishing the normal respiration from the hypopnea, finally completing the overall classification. Refer to Fig. 2 for the overall flowchart. Compared with prior art, the SpCh signals used in the embodiments of the present invention may be obtained from a medical-grade finger pulse oxygen sensor, or from cheap wearable devices such as smart bracelets, and only require an ultra-low signal resolution of 1 Hz. Low-resolution signals (sampling rate equal 10 to 1 Hz) may be used directly, and high-resolution signals (sampling rate greater than 1 Hz) may be down-sampled for use. Therefore, the present invention provides a cheaper, more flexible and comfortable detection method while maintaining high detection accuracy. The embodiments of the present invention provide an OSAHS detection system, comprising: a first module for obtaining original blood oxygen signal data, preprocessing the original blood oxygen ^5 signal data to obtain blood oxygen signal data slice; a second module for analyzing the blood oxygen C\j signal data slice by using a one-dimensional neural network classifier to obtain first-stage LO classification results, wherein the one-dimensional neural network classifier is generated by training blood oxygen signal data of a labelled sleep-related respiratory event, and the first-stage classification results include normal respiration, suspected hypopnea, and apnea; a third module for analyzing the OJ blood oxygen signal data slice of a target fragment by using the classifier based on rule matching to obtain second-stage classification results, wherein the classifier based on rule matching is generated using scores of blood oxygen signal statistical indexes as discrimination criteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage 25 classification results comprise normal respiration, hypopnea and apnea. In a further aspect, the embodiments of the present invention further provide an OSAHS detection device, comprising a processor and a memory. The memory is configured to store programs. The processor is configured to execute the programs to implement the method as described 30 herein. The contents of the system embodiments of the present invention are all applicable to the device embodiments, and the functions thereof specifically implemented by the device embodiments are the same as those of the above system embodiments, and the beneficial effects achieved thereby are also the same as those achieved by the above system. In a still further aspect, the embodiments of the present invention further provide a computer-readable storage medium storing programs therein, wherein the programs are executed by a processor to implement the method as described herein. 5 The contents of the system embodiments of the present invention are all applicable to the embodiments of the computer-readable storage medium, and the functions thereof specifically implemented by the embodiments of the computer-readable storage medium are the same as those of the above system embodiments, and the beneficial effects achieved thereby are also the same as those achieved by the method as described herein. 10 The embodiments of the present invention further disclose a computer program product or a computer program, wherein the computer program product or the computer program comprises computer instructions that are stored in a computer-readable storage medium. A processor of a computer device may read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions such that the computer device executes the method as described herein. r\l M In some alternative embodiments, the functions / operations mentioned in block diagrams may LO occur out of the order noted in an operation diagram. For example, two blocks shown in succession may actually be executed substantially concurrently or the blocks may sometimes be executed in a reverse order, depending on the functions / operations involved. Furthermore, the embodiments 20 presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more comprehensive understanding of this technology. The disclosed method is not limited to the operations and logical flow presented herein. Alternative embodiments are predictable, in which the order of various operations is changed and sub-operations described as a part of larger operations are performed independently. 25 Furthermore, although the present invention is described in the context of functional modules, it should be understood that unless stated to the contrary, one or more of the described functions and / or features may be integrated into a single physical device and / or software modules, or one or more functions and / or features may be implemented in separate physical devices or software modules. It may also be understood that a detailed discussion of the actual implementation of each module is not 30 essential to understand the present invention. Or rather, given the attributes, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the modules will be understood in the conventional techniques of engineers. Therefore, those having ordinary skills in the art can implement the present invention set forth in the claims using ordinary techniques, without undue experimentation. It may also be understood that the particular concepts disclosed are illustrative only and are not intended to limit the scope of protection of the invention, which is to be determined by the full scope of the appended claims. When the functions are implemented in the form of software functional units and sold or used 5 as independent products, the functions may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the invention, in essence, or its contribution to the prior art, or part of the technical solution may be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to 10 perform all or some of the steps of the method in various embodiments of the invention. The foregoing storage medium includes: any medium that can store program code, such as a USB flash drive, a removable hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc. The logics and / or steps represented in the flowchart or described in other manners herein can, ^5 for example, be considered as a sequence table of executable instructions for implementing logical C\j functions, which can be embodied in any computer-readable medium to be used by an instruction LO execution device, apparatus or equipment (such as a computer-based device, a device including a C* ^ processor, or other devices capable of extracting instructions from an instruction execution device, apparatus or equipment and executing the instructions), or to be used in conjunction with such an OJ instruction execution device, apparatus or equipment. For this specification, the term "computer-readable medium" may be any device that may contain, store, communicate with, propagate or transmit a program to be used by or used in conjunction with an instruction execution device, apparatus or equipment. More specific examples (non-exhaustive list) of the computer-readable media include the 25 following: an electrical connection portion (electronic device) with one or more wires, a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable-programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable media on which the program may be printed, because 30 the program may be obtained electronically and then stored in the computer memory, for example, by optically scanning the paper or other media, then performing editing, interpretation or processing in other suitable manners if necessary. It should be understood that various parts of the present invention may be implemented by hardware, software, firmware or a combination thereof. In the above-described embodiments, the 90 multiple steps or methods may be implemented by using software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if the steps or methods are implemented by using the hardware, they are the same as those in another embodiment and may be implemented by any one or a combination of the following techniques known in the art: a discrete 5 logic circuit with a logic gate circuit for achieving a logic function to data signals, an application specific integrated circuit with a suitable combinational logic gate, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. In the description of this specification, descriptions with reference to the term such as "an embodiment", "some embodiments", "example", "specific example", or "some examples" mean that 10 specific features, structures, materials, or characteristics described with reference to the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiments or examples. Moreover, the specific features, structures, materials, or characteristics described herein may be combined in any one or more embodiments or examples in a suitable manner. Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art may appreciate that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principle and gist of the present invention, and the scope of the present invention is defined by the appended claims. The foregoing provides specific description of the preferred embodiments of the present invention, but the present invention is not limited to the above-described embodiments The scope of the present invention is as defined in the appended claims.

Claims

1. An OSAHS detection system, comprising:a first module for obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice;a second module for analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results; wherein the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; anda third module for analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discrimination criteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage classification results include normal respiration, hypopnea and apnea.

2. The OSAHS detection system according to claim 1, wherein obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice comprises:obtaining original blood oxygen signal data of a target object in a preset time period;screening, based on a preset proportion threshold of a blood oxygen artifact, the original blood oxygen signal data to obtain target blood oxygen signal data;slicing the target blood oxygen signal data based on a preset duration; anddown-sampling and normalizing the target blood oxygen signal data obtained after the slicing, to obtain blood oxygen signal data slice.

3. The OSAHS detection system according to claim 1, wherein the system is for:creating a one-dimensional neural network classifier based on the blood oxygen signal data of the labelled sleep-related respiratory event.

4. The OSAHS detection system according to claim 3, wherein creating one-dimensional neural network classifier based on blood oxygen signal data of the labelled sleep-related respiratory event comprises:determining training samples based on the blood oxygen signal data of the labelled sleep-related respiratory event;providing a neural network model with two branches based on one-dimensional convolution operator, a one-dimensional maximum pooling operator, an activation function and a fully connected layer, performing classification training on the neural network model based on the training samples, and adjusting the neural network model based on training results to obtain a one-dimensional neural5 network classifier, wherein the two branches of the neural network model each comprise several onedimensional convolution operators, a one-dimensional maximum pooling operator and an activation function, and the two branches of the neural network model are connected in the fully connected layer.

5. The OSAHS detection system according to claim 1, wherein the one-dimensional neural 10 network classifier comprises two branches, and analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results comprises:performing fragment extension according to each fragment in the blood oxygen signal data slice to obtain a first branch input; and performing fragment combination to obtain a second branch input;performing convolution operation, activation response and dimensionality reduction respectively on the first branch input and the second branch input to correspondingly obtain a first C\l feature map and a second feature map;LO splicing the first feature map and the second feature map according to a channel dimension to obtain a third feature map; and1—performing fully connected classification on the third feature map to obtain the first-stage 20 classification results.

6. The OSAHS detection system according to claim 1, wherein the system is for:setting discrimination rules for the scores of the blood oxygen signal statistical indexes based on blood oxygen saturations, blood oxygen signal variances and blood oxygen minimum values; andcreating a classifier based on rule matching according to the discrimination rules.25 7. The OSAHS detection system according to claim 6, wherein analyzing the blood oxygensignal data slice of a target fragment by using a classifier based on rule matching comprises:performing first scoring processing according to a reduction amplitude of the blood oxygen saturation of the blood oxygen signal data slice of the target fragment;performing second scoring processing according to a fluctuation range of the blood oxygen30 signal variance of the blood oxygen signal data slice of the target fragment;performing third scoring processing according to a reduction amplitude of the blood oxygen minimum value of the blood oxygen signal data slice of the target fragment;21 05 24determining the scores of the blood oxygen signal statistical indexes by means of the first scoring processing, the second scoring processing and the third scoring processing; andobtaining second-stage classification results according to the scores of the blood oxygen signal statistical indexes.5 8. An OSAHS detection device, comprising: a processor and a memory;the memory being configured to store a program; andthe processor executing the program to implement a method comprising:obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice;10 analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results; wherein the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; andanalyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discrimination criteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the20 second-stage classification results include normal respiration, hypopnea and apnea.

9. A computer-readable storage medium, wherein the computer-readable storage medium stores a program, and the program is executed by a processor to implement a method comprising:obtaining original blood oxygen signal data, and preprocessing the original blood oxygen signal data to obtain blood oxygen signal data slice;25 analyzing the blood oxygen signal data slice by using a one-dimensional neural network classifier to obtain first-stage classification results; wherein the one-dimensional neural network classifier is generated by training the blood oxygen signal data of a labelled sleep-related respiratory event; the first-stage classification results include normal respiration, suspected hypopnea, and apnea; and30 analyzing the blood oxygen signal data slice of a target fragment by using a classifier based on rule matching, to obtain second-stage classification results; wherein the classifier based on rule matching is generated by using scores of blood oxygen signal statistical indexes as discriminationcriteria; the target fragment comprises fragments corresponding to the blood oxygen signal data slice with the first-stage classification results that are normal respiration and suspected hypopnea; the second-stage classification results include normal respiration, hypopnea and apnea.21 05 24

Images