An OSAHS detection method, system, device and medium

Abstract

The present invention discloses an OSAHS detection method, system, device and medium. The method includes: obtaining original blood oxygen signal data, performing data preprocessing on the original blood oxygen signal data to obtain blood oxygen signal slice data; using a one-dimensional neural network classifier to analyze the blood oxygen signal slice data to obtain a first-stage classification result; using a classifier based on rule matching to analyze the blood oxygen signal slice data of a target segment to obtain a second-stage classification result. The present invention can achieve the simplicity of signal acquisition and effectively reduce the detection cost only based on the blood oxygen signal. At the same time, the first-stage classification is realized through a learning-based neural network to ensure the generalization of the classification result, and then the second-stage classification is realized by combining the scores of blood oxygen signal statistical indicators to achieve classification refinement and deviation correction, and obtain an accurate classification result. The present invention can achieve accurate and low-cost OSAHS detection and can be widely applied to the field of data detection technology.

Description

Technical Field

[0001] The present invention relates to the technical field of data detection, and in particular to an OSAHS detection method, system, device and medium. Background Art

[0002] Obstructive sleep apnea hypopnea syndrome (OSAHS) is the most common sleep breathing disorder at present. Its characteristic is that the upper respiratory tract intermittently partially or completely collapses during sleep, causing repeated hypoventilation (hypopnea) and apnea, resulting in frequent drops in blood oxygen saturation, hypercapnia and sleep fragmentation. With the development of obesity and aging, the prevalence of OSAHS has increased rapidly. It is estimated that 936 million people aged 30 to 69 worldwide suffer from this disease. OSAHS can cause acute respiratory failure and sudden death, increase the prevalence of hypertension, stroke, coronary heart disease, Alzheimer's disease and diabetes, increase the mortality rate of patients with chronic obstructive pulmonary disease, renal insufficiency and heart failure, and has become a global health problem.

[0003] Polysomnography (PSG), first reported in 1974, is still the gold standard for diagnosing OSAHS. That is, in a sleep monitoring room, a polysomnograph is used to continuously and synchronously collect and record multiple sleep physiological parameters, including electroencephalogram, electrooculogram, electromyogram, electrocardiogram, oronasal airflow, pulse oxygen saturation (SpO2), thoracoabdominal respiratory movement, body position and snoring, etc.; sleep experts interpret and analyze these parameters according to the latest version of the "American Academy of Sleep Medicine (AASM) Manual for the Scoring of Sleep and Associated Events: Rules, Terminology, and Technical Specifications", and finally obtain the apnea-hypopnea index (AHI) used to describe the frequency of sleep breathing events, and use it as an index for diagnosing OSAHS and classifying the severity of the disease [7]: AHI < 5 is non-OSAHS, 5 ≤ AHI < 15 is mild, 15 ≤ AHI < 30 is moderate, and AHI ≥ 30 is severe. However, the above entire process has a high technical threshold, a large burden on patients, and is expensive. It is both time-consuming and laborious and prone to errors, significantly affecting the practicality and accessibility of OSAHS diagnostic resources.

[0004] According to the AASM recommended rules, an apnea event is defined as a ≥90% decrease in respiratory airflow from the baseline value for at least 10 s; a hypopnea event is defined as a ≥30% decrease in respiratory airflow from the baseline value for at least 10 s, accompanied by a ≥3% decrease in blood oxygen saturation or arousal; and the AHI is defined as the sum of the number of apnea and hypopnea events per hour of total sleep time, which is an important indicator for conveying information on sleep-disordered breathing. It is worth mentioning that the initial research on respiratory events only focused on apnea. However, it was soon found that the impairment of the body caused by weakened airflow is as severe as that of apnea and also has clinical significance: on the one hand, identifying hypopnea events is related to the accurate calculation of the AHI value, which directly affects the diagnosis and treatment decisions of OSAHS. On the other hand, in some OSAHS patients, the main respiratory events recorded by PSG are not apnea but hypopnea. Identifying hypopnea events is beneficial for individualized diagnosis and treatment. Studies have confirmed that extremely obese subjects (BMI ≥ 45 kg / m2) with OSAHS exhibit a very large number of hypopneas and relatively few apneas, which supports the possibility that the generation mechanisms of hypopnea and apnea may be different and may lead to specific treatments for sleep-disordered breathing dominated by hypopnea.

[0005] However, at the present stage, most studies only identify apnea and ignore hypopnea events, unable to comprehensively and accurately describe sleep-disordered breathing, which is likely to cause underestimation or even missed diagnosis of the condition.

[0006] In view of this, how to achieve accurate and low-cost detection of OSAHS is an urgent problem to be solved. Summary of the Invention

[0007] In view of this, the embodiments of the present invention provide an OSAHS detection method, system, device and medium, which can achieve accurate and low-cost detection of OSAHS.

[0008] On the one hand, the embodiments of the present invention provide an OSAHS detection method, including:

[0009] Obtain the original blood oxygen signal data, perform data preprocessing on the original blood oxygen signal data to obtain blood oxygen signal slice data;

[0010] Use a one-dimensional neural network classifier to analyze the blood oxygen signal slice data to obtain a first-stage classification result; wherein, the one-dimensional neural network classifier is trained and generated by the blood oxygen signal data with labeled sleep respiratory events; the first-stage classification result includes normal breathing, suspected hypopnea and apnea;

[0011] Using a classifier based on rule matching, analyze the slice data of the blood oxygen signal of the target segment to obtain a two-stage classification result; wherein, the classifier based on rule matching is generated by using the score of the blood oxygen signal statistical index as the discrimination criterion; the target segment includes the segments corresponding to the slice data of the blood oxygen signal with the one-stage classification results of normal breathing and suspected hypoventilation; the two-stage classification results include normal breathing, hypoventilation, and apnea.

[0012] Optionally, obtain the original blood oxygen signal data, perform data preprocessing on the original blood oxygen signal data to obtain the slice data of the blood oxygen signal, including:

[0013] Obtain the original blood oxygen signal data of the target object during a preset time period;

[0014] Based on a preset proportional threshold of blood oxygen artifacts, screen the original blood oxygen signal data to obtain the target blood oxygen signal data;

[0015] Based on a preset duration, perform data slicing on the target blood oxygen signal data;

[0016] Perform downsampling and normalization on the sliced target blood oxygen signal data to obtain the slice data of the blood oxygen signal.

[0017] Optionally, it further includes:

[0018] Create a one-dimensional neural network classifier based on the blood oxygen signal data of the labeled sleep breathing events.

[0019] Optionally, creating a one-dimensional neural network classifier based on the blood oxygen signal data of the labeled sleep breathing events includes:

[0020] Determine the training samples according to the blood oxygen signal data of the labeled sleep breathing events;

[0021] Based on a one-dimensional convolutional operator, a one-dimensional max pooling operator, an activation function, and a fully connected layer, set up a neural network model with two branches, perform classification training on the neural network model based on the training samples, and adjust the neural network model based on the training results to obtain a one-dimensional neural network classifier; wherein, both branches of the neural network model include a number of one-dimensional convolutional operators, one-dimensional max pooling operators, and activation functions; the two branches of the neural network model are connected at the fully connected layer.

[0022] Optionally, the one-dimensional neural network classifier includes two branches. Using the one-dimensional neural network classifier, analyze the slice data of the blood oxygen signal to obtain the one-stage classification result, including:

[0023] According to each segment in the slice data of the blood oxygen signal, perform segment extension to obtain the input of the first branch; and perform segment combination to obtain the input of the second branch;

[0024] Perform convolution operations, activation responses, and dimensionality reduction on the first branch input and the second branch input respectively, and correspondingly obtain a first feature map and a second feature map;

[0025] Concatenate the first feature map and the second feature map along the channel dimension to obtain a third feature map;

[0026] Perform fully connected classification on the third feature map to obtain a first-stage classification result.

[0027] Optionally, it further includes:

[0028] Set the discrimination rules for the scores of the blood oxygen signal statistical indicators based on the blood oxygen saturation, the variance of the blood oxygen signal, and the minimum blood oxygen value;

[0029] Create a classifier based on rule matching according to the discrimination rules.

[0030] Optionally, use the classifier based on rule matching to analyze the blood oxygen signal slice data of the target segment to obtain a second-stage classification result, including:

[0031] Perform a first score assignment process according to the decrease amplitude of the blood oxygen saturation of the blood oxygen signal slice data of the target segment;

[0032] Perform a second score assignment process according to the fluctuation amplitude of the variance of the blood oxygen signal of the blood oxygen signal slice data of the target segment;

[0033] Perform a third score assignment process according to the decrease amplitude of the minimum blood oxygen value of the blood oxygen signal slice data of the target segment;

[0034] Determine the score of the blood oxygen signal statistical indicator based on the first score assignment process, the second score assignment process, and the third score assignment process;

[0035] Obtain a second-stage classification result according to the score of the blood oxygen signal statistical indicator.

[0036] On the other hand, an embodiment of the present invention provides an OSAHS detection system, including:

[0037] A first module, configured to obtain the original blood oxygen signal data, perform data preprocessing on the original blood oxygen signal data, and obtain blood oxygen signal slice data;

[0038] A second module, configured to use a one-dimensional neural network classifier to analyze the blood oxygen signal slice data to obtain a first-stage classification result; wherein, the one-dimensional neural network classifier is trained and generated by the blood oxygen signal data with labeled sleep apnea events; the first-stage classification result includes normal breathing, suspected hypoventilation, and apnea;

[0039] A third module is used to analyze the slice data of the blood oxygen signal of the target segment by using a classifier based on rule matching to obtain a two-stage classification result. The classifier based on rule matching is generated by using the scores of blood oxygen signal statistical indicators as the discrimination criteria. The target segment includes the segments corresponding to the slice data of the blood oxygen signal with the first-stage classification results of normal breathing and suspected hypopnea. The two-stage classification results include normal breathing, hypopnea, and apnea.

[0040] On the other hand, an embodiment of the present invention provides an OSAHS detection device, including a processor and a memory.

[0041] The memory is used to store programs.

[0042] The processor executes the program to implement the method as described above.

[0043] On the other hand, an embodiment of the present invention provides a computer-readable storage medium, and the storage medium stores a program, and the program is executed by the processor to implement the method as described above.

[0044] An embodiment of the present invention also discloses a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to execute the method as described above.

[0045] In the embodiment of the present invention, the original blood oxygen signal data is first obtained, and the original blood oxygen signal data is preprocessed to obtain the slice data of the blood oxygen signal. A one-dimensional neural network classifier is used to analyze the slice data of the blood oxygen signal to obtain a first-stage classification result. The one-dimensional neural network classifier is trained and generated by using the blood oxygen signal data with labeled sleep breathing events. The first-stage classification results include normal breathing, suspected hypopnea, and apnea. A classifier based on rule matching is used to analyze the slice data of the blood oxygen signal of the target segment to obtain a two-stage classification result. The classifier based on rule matching is generated by using the scores of blood oxygen signal statistical indicators as the discrimination criteria. The target segment includes the segments corresponding to the slice data of the blood oxygen signal with the first-stage classification results of normal breathing and suspected hypopnea. The two-stage classification results include normal breathing, hypopnea, and apnea. The detection data of the present invention is only based on the blood oxygen signal, which can realize the simplicity of signal acquisition and effectively reduce the detection cost. At the same time, based on the one-dimensional neural network classifier and the classifier based on rule matching, the first-stage classification is first realized through the learning-based neural network to ensure the generalization of the classification result, and then the two-stage classification is realized by combining the scores of the blood oxygen signal statistical indicators to realize classification refinement and deviation correction, and an accurate classification result is obtained. The present invention can realize accurate and low-cost OSAHS detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0047] Figure 1 It is a schematic flow chart of an OSAHS detection method provided by an embodiment of the present invention;

[0048] Figure 2 It is a schematic overall flow chart of an OSAHS detection method provided by an embodiment of the present invention;

[0049] Figure 3 It is a schematic structural diagram of a one-dimensional neural network classifier provided by an embodiment of the present invention;

[0050] Figure 4 It is a schematic diagram of the blood oxygen variance in different states provided by an embodiment of the present invention;

[0051] Figure 5 It is a schematic diagram of the minimum blood oxygen value in different states provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] First of all, it should be noted that at the present stage, for OSAH detectors based on blood oxygen, some scholars classify hypopnea events and apnea events into one category to reduce the complexity of algorithm implementation and improve the accuracy. However, confusing apnea and hypopnea makes it impossible to distinguish whether a patient's oxygen drop at a certain time is caused by hypopnea or apnea.

[0054] And some scholars have tried to use different machine learning methodologies to distinguish between apnea and hypopnea. There is a method in the prior art that uses sparse representation structured dictionary learning for OSAH detection, but the effect is not good. On the SHHS dataset (The Sleep Heart Health Study database), the hypopnea sensitivity is only 22.78%.

[0055] Some scholars also participate in the detection by relying on the gold standard for OSAH diagnosis, that is, using airflow signals to assist in identifying hypopnea. There is an existing technology that uses a two-dimensional convolutional neural network (2DCNN) combined with a fully connected neural network (FCN) to design an OSAH detector, which uses the combination of airflow and blood oxygen signals to detect childhood OSAH and has achieved good results in the CHAT dataset (the Childhood Adenotonsillectomy Trial database). However, it should be noted that the cost of obtaining airflow signals is more expensive than that of blood oxygen signals, with a cost difference of about ten times.

[0056] Different from the above method, the present invention proposes a two-stage classification model based on single-channel blood oxygen signals, which realizes the automatic and accurate detection of two respiratory events, apnea and hypopnea: (1) The input of the model is only the blood oxygen signal of the subject, considering the simplicity of signal acquisition and patient comfort; (2) The two-stage model combines neural networks with blood oxygen time-domain statistical information, which can improve the recognition rate of the algorithm for hypopnea, and has a low training complexity, and the inference process can be completed on low-power devices such as smart bracelets.

[0057] On the one hand, referring to Figure 1 , an embodiment of the present invention provides an OSAHS detection method, including:

[0058] S100. Obtain the original blood oxygen signal data, perform data preprocessing on the original blood oxygen signal data, and obtain blood oxygen signal slice data;

[0059] It should be noted that in some embodiments, the original blood oxygen signal data of the target object in a preset time period is obtained; based on a preset proportional threshold of blood oxygen artifacts, the original blood oxygen signal data is screened to obtain target blood oxygen signal data; based on a preset duration, the target blood oxygen signal data is sliced; the downsampling and normalization are performed on the sliced target blood oxygen signal data to obtain blood oxygen signal slice data.

[0060] Specifically, first, to ensure that the SpO 2 signal (i.e., blood oxygen signal) is collected from the subject's sleep state, the embodiment of the present invention uses the data intercepted from 0:00 midnight to 6:00 am as a better time period. Second, to reduce the interference of noise signals, only the data with the proportion of blood oxygen artifacts (artifact) in the total recording length <1% is selected, and the remaining blood oxygen artifacts are eliminated using nearest neighbor interpolation. Third, the data is sliced according to a 1-minute length. Finally, the dataset SpO 2The sampling frequency of the signal is 10 Hz. The blood oxygen time series is downsampled from 10 Hz to 1 Hz by the open-source largest triangle three buckets (LTTB) downsampling algorithm, and Z-Score normalization is performed, that is, all SpO 2 data values are subtracted by their mean and then divided by the variance.

[0061] S200. Use a one-dimensional neural network classifier to analyze the sliced blood oxygen signal data to obtain a first-stage classification result;

[0062] It should be noted that the one-dimensional neural network classifier is generated by training with the blood oxygen signal data of the labeled sleep apnea events; the first-stage classification results include normal breathing, suspected hypoventilation, and apnea;

[0063] In some embodiments, it further includes: creating a one-dimensional neural network classifier based on the blood oxygen signal data of the labeled sleep apnea events; wherein, according to the blood oxygen signal data of the labeled sleep apnea events, training samples are determined; based on one-dimensional convolutional operators, one-dimensional max-pooling operators, activation functions, and fully connected layers, a neural network model with two branches is set up, the neural network model is classified and trained based on the training samples, and based on the training results, the neural network model is adjusted to obtain a one-dimensional neural network classifier; wherein, both branches of the neural network model include several one-dimensional convolutional operators, one-dimensional max-pooling operators, and activation functions; the two branches of the neural network model are connected at the fully connected layer.

[0064] Specifically, as Figure 2 shown, the dataset used in the present invention comes from the OSAHS clinical research integration big data platform of the First Affiliated Hospital of Sun Yat-sen University. Only the SpO 2 single-channel data is used for the training, verification, and testing of the algorithm. A total of 450 subject data are used. After data cleaning, 358 pieces of data are available. They are divided into a training set (i.e., training samples), a verification set, and a test set according to a ratio of 6:2:2, that is, 218 pieces of data in the training set, 70 pieces of data in the verification set, and 70 pieces of data in the test set. Due to poor pick-up quality, the patient's sweating or turning over at night and other actions may cause the sensor to be poorly contacted or even fall off, resulting in SpO 2Signal interruption, this situation is called artifact. When the length of the artifact exceeds 10% of the overnight signal, this record is deprecated; when the length of the artifact is less than 10%, the nearest neighbor linear interpolation is used to erase the artifact, and the whole process is called data cleaning. Among them, the training set will also go through the data preprocessing in the specific embodiments of step S100. In the third step, respiratory event markers are made for each 1-minute length segment after data slicing. These segments are marked as normal breathing (normal, N), apnea (apnea, A), and hypopnea (hypopnea, H). Then, based on the preprocessed training set, the neural network model is trained and adjusted, and finally a one-dimensional neural network classifier is obtained. Among them, the input of the neural network model has two branches, and the input signals are all blood oxygen signals downsampled to 1 Hz. The detailed structure is as Figure 3 shown. In some specific embodiments, Flatten is used at the ends of both branches as a method to flatten the feature map.

[0065] In some embodiments, using the one-dimensional neural network classifier, the blood oxygen signal slice data is analyzed to obtain a first-stage classification result, including: extending the segments according to the segments in the blood oxygen signal slice data to obtain the input of the first branch; and combining the segments to obtain the input of the second branch; performing convolution operations, activation responses, and dimensionality reduction on the input of the first branch and the input of the second branch respectively to obtain the first feature map and the second feature map; splicing the first feature map and the second feature map according to the channel dimension to obtain the third feature map; performing fully connected classification on the third feature map to obtain the first-stage classification result.

[0066] Specifically, as Figure 3 shown:

[0067] (1) The first input branch is denoted as Input1 Spo2 . Generally, during sleep, the blood oxygen saturation shows a downward trend with the increase of the respiratory event time, but not every time the two occur simultaneously. Considering that there is a delay of 1 to 10 seconds between the oxygen desaturation (OD) and the respiratory event, and the length varies from person to person. To offset the delay, the input length is extended from the standard 1 minute to 1 minute and 20 seconds (i.e., segment extension), totaling 80 seconds. The 20 seconds extended in the current minute are the first 20 seconds of the next minute, that is, there is a 20-second signal window overlap, so that the blood oxygen drop signal caused by the sleep respiratory event occurring at the rear of the current minute detection window can also be captured. The blood oxygen signal segment then passes through three one-dimensional convolutional operators (Conv1D), which are designed to extract useful features from the input and save the computational cost of feature extraction by sharing weights and receptive fields through convolutional operations. The one-dimensional feature map extracted by the one-dimensional convolutional operator is denoted as F 1 , and the weight of the operator is denoted as θ 1ZSCORE normalization can produce negative values. To ensure the neural network's response ability to negative values, LeakyReLU is used here as the activation function, which can also respond to negative values, different from ReLU that completely does not respond to negative values. And max pooling (Maxpooling1D) is used for dimensionality reduction to further reduce the number of parameters in the neural network, so as to reduce the overhead of forward inference. The data expression of the first branch is as follows:

[0068] F 1 = f(Input1 SpO2 , θ 1 )

[0069] (2) The second input branch is denoted as Input2 SpO2 . For patients with relatively severe OSAHS, their sleep apnea events occur frequently and are often concentrated. For example, if a sleep apnea event occurs within the current 1 minute, there is a high probability that it will occur again in the next 1 minute, that is, the occurrence of sleep apnea events has quasi-periodicity. To capture this quasi-periodicity and improve the prediction accuracy, a second input branch is designed, which will read the blood oxygen signals within a relatively large range of 300 seconds (i.e., fragment combination) of the current 1 minute and 2 minutes before and after it. Similar to the first input branch, it also passes through a one-dimensional convolutional neural network, and the one-dimensional feature map extracted by its one-dimensional convolutional operator is denoted as F 2 , and the weight of the operator is denoted as θ 2 . The data expression of the second branch is as follows:

[0070] F 2 = f(Input2 Spo2 , θ 1 )

[0071] Then, the feature maps F 1 and F 2 extracted by the two branches are concatenated along the channel dimension and sent into the fully connected classification network D apnea (including the Dense fully connected layer) to complete the first-stage classification. D 3 is the weight of the fully connected classification network. The output of the one-dimensional neural network classifier is normal breathing and suspected hypopnea (N&H) or apnea (A). Normal breathing and suspected hypopnea (N&H) will be further classified in the second stage to distinguish normal breathing segments from suspected hypopnea segments. The data expression is as follows:

[0072] F concat = f(F 1 , F 2 )

[0073] D apnea = f(F concat , θ 3)

[0074] S300. Analyze the slice data of the blood oxygen signal of the target segment using a classifier based on rule matching to obtain a two-stage classification result;

[0075] It should be noted that the classifier based on rule matching is generated by using the score of the blood oxygen signal statistical index as the discrimination criterion; the target segment includes the segments corresponding to the slice data of the blood oxygen signal with the first-stage classification results of normal breathing and suspected hypopnea; the two-stage classification results include normal breathing, hypopnea, and apnea;

[0076] In some embodiments, it further includes: setting a discrimination rule for the score of the blood oxygen signal statistical index based on blood oxygen saturation, variance of the blood oxygen signal, and minimum blood oxygen value; creating a classifier based on rule matching according to the discrimination rule.

[0077] Among them, analyzing the slice data of the blood oxygen signal of the target segment using a classifier based on rule matching to obtain a two-stage classification result includes: performing a first score assignment process according to the decrease amplitude of the blood oxygen saturation of the slice data of the blood oxygen signal of the target segment; performing a second score assignment process according to the fluctuation amplitude of the variance of the blood oxygen signal of the slice data of the blood oxygen signal of the target segment; performing a third score assignment process according to the decrease amplitude of the minimum blood oxygen value of the slice data of the blood oxygen signal of the target segment; determining the score of the blood oxygen signal statistical index based on the first score assignment process, the second score assignment process, and the third score assignment process; obtaining a two-stage classification result according to the score of the blood oxygen signal statistical index.

[0078] Specifically, for some hypopnea events, the peak value of the airflow curve decreases less than the baseline value, and the resulting decrease in blood oxygen is not obvious, and its characteristics are difficult to be captured by the neural network. Therefore, the embodiment of the present invention introduces a second-stage classifier based on rule matching to identify hypopnea respiratory events. This classifier uses the score of the blood oxygen signal statistical index as the discrimination criterion: the blood oxygen signal segment that meets the statistical index will be scored, and when the score reaches a certain level, the segment will be determined as a hypopnea event. Three scoring criteria will be introduced below:

[0079] (1) 3% OD. When the blood oxygen saturation decreases by ≥ 3% compared to the baseline value, assign 1 point, otherwise do not assign points (i.e., the first score assignment process), which is also one of the recommended rules for judging hypopnea in the "AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology, and Technical Specifications".

[0080] (2) Variance of blood oxygen signal. Variance is a very good indicator for measuring signal fluctuations. Whether in a waking or sleeping state, the blood oxygen fluctuations of normal people are extremely small. However, when a sleep breathing event occurs, due to airway collapse leading to a reduction in inhaled air flow, OSAHS patients are unable to intake sufficient oxygen, and the blood oxygen level immediately drops and deviates from the normal value, thereby increasing the variance. As Figure 4 shown in the display of the blood oxygen variance in different states, when the variance > 0.5, 1 point is assigned, which is also the upper quartile line of the normal breathing state and the lower quartile line of the hypopnea state; when the variance > 9, 2 points are assigned, which is also the upper limit of the blood oxygen variance during hypopnea, indicating that a more serious breathing event than hypopnea may occur at this time, that is, apnea; and when the variance < 0.5, no points are assigned. The above represents the details of the second point assignment process.

[0081] (3) Minimum blood oxygen value. As Figure 5 shown in the display of the minimum blood oxygen value in different states, for normal people, a blood oxygen level less than 94% is considered low blood oxygen. However, in the embodiments of the present invention, it is desired to distinguish normal segments from hypopnea segments as much as possible. Therefore, when the minimum blood oxygen value is less than 90%, 1 point is assigned, otherwise no points are assigned (i.e., the third point assignment process).

[0082] When the total score ≥ 2, it indicates that this blood oxygen segment is very likely a hypopnea event; when the total score reaches 4, it indicates that this blood oxygen segment is very likely an apnea event, but it has not been marked as such by the first-stage neural network and needs to be re-marked as apnea for this segment, otherwise, it is normal breathing.

[0083] Among them, the OSAH detection and evaluation indicators mainly include accuracy (Acc), sensitivity (Sens), specificity (Spec), how many of the samples predicted to be true are positive samples (Prec), and F 1 value, and the calculation methods are as follows:

[0084]

[0085]

[0086]

[0087]

[0088]

[0089] Among them, TP, TN, FP, and FN represent "true positive", "true negative", "false positive", and "false negative" respectively. In a multi-classification task (referring to the number of classes to be classified > 3), the embodiments of the present invention will measure each index in the form of macro-average, that is, calculate the respective indexes for each category and then calculate the average value.

[0090] In some specific embodiments, the present invention conducts experimental verification on the effectiveness of the above method. Specifically, the proposed algorithm is tested on the test set of the OSAH dataset provided by the First Affiliated Hospital of Sun Yat-sen University. We ensure that the test data does not participate in the training and verification phases of the algorithm and is only used for testing the two-stage model. Our test results are shown in Table 1, and the accuracy can reach 80.28%.

[0091] Table 1

[0092] Accuracy (%) Sensitivity (%) Specificity (%) F1 Score 80.28 75.41 90.84 0.8028

[0093] It is true that at the present stage, the accuracy of using airflow signals to detect OSAHS has reached over 90%. However, the airflow signals not only have a high acquisition cost, that is, a thermal sensor and a disposable pressure airflow tube are required, but also seriously affect the patient's sleep quality. The thermal sensor and the pressure airflow tube need to be placed at the patient's mouth and nose and inserted into the nasal cavity, resulting in extremely poor comfort. And the SpO 2 signal used in the embodiments of the present invention can be taken from a medical-grade finger pulse oximeter sensor or can also be sourced from wearable devices such as smart watches and inexpensive smart bracelets. Moreover, only an ultra-low signal resolution of 1 Hz is required. Low-resolution signals (sampling rate = 1 Hz) can be directly used, and high-resolution signals (sampling rate > 1 Hz) can be downsampled for use. Therefore, the present invention provides a more inexpensive, flexible, and comfortable detection method while maintaining a high detection accuracy.

[0094] In summary, the present invention conducts classification detection in two stages. In the first stage, a one-dimensional neural network classifier is used for preliminary classification to distinguish apnea data from other label data. In the second stage, based on rule matching, the normal breathing and hypopnea are distinguished by using the statistical information of the blood oxygen time series, and the overall classification is finally completed. The overall flowchart is referred to Figure 2 . Compared with the prior art, the SpO2 signal used in the embodiments of the present invention can be taken from a medical-grade finger pulse oximeter sensor or can also be sourced from wearable devices such as smart watches and inexpensive smart bracelets. Moreover, only an ultra-low signal resolution of 1 Hz is required. Low-resolution signals (sampling rate = 1 Hz) can be directly used, and high-resolution signals (sampling rate > 1 Hz) can be downsampled for use. Therefore, the present invention provides a more inexpensive, flexible, and comfortable detection method while maintaining a high detection accuracy.

[0095] On the other hand, an embodiment of the present invention provides an OSAHS detection system, including: a first module for acquiring original blood oxygen signal data, performing data preprocessing on the original blood oxygen signal data to obtain blood oxygen signal slice data; a second module for analyzing the blood oxygen signal slice data by using a one-dimensional neural network classifier to obtain a first-stage classification result; wherein, the one-dimensional neural network classifier is trained and generated by using the blood oxygen signal data with labeled sleep apnea events; the first-stage classification result includes normal breathing, suspected hypopnea, and apnea; a third module for analyzing the blood oxygen signal slice data of a target segment by using a classifier based on rule matching to obtain a second-stage classification result; wherein, the classifier based on rule matching is generated by using the score of the blood oxygen signal statistical index as a discrimination criterion; the target segment includes the segments corresponding to the blood oxygen signal slice data with the first-stage classification results of normal breathing and suspected hypopnea; the second-stage classification result includes normal breathing, hypopnea, and apnea.

[0096] The content of the method embodiment of the present invention is applicable to the system embodiment of the present invention. The functions specifically implemented by the system embodiment of the present invention are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those of the above method.

[0097] Another aspect of the embodiment of the present invention further provides an OSAHS detection device, including a processor and a memory;

[0098] The memory is used to store a program;

[0099] The processor executes the program to implement the method as described above.

[0100] The content of the method embodiment of the present invention is applicable to the device embodiment of the present invention. The functions specifically implemented by the device embodiment of the present invention are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those of the above method.

[0101] Another aspect of the embodiment of the present invention further provides a computer-readable storage medium, where the storage medium stores a program, and the program is executed by a processor to implement the method as described above.

[0102] The content of the method embodiment of the present invention is applicable to the computer-readable storage medium embodiment of the present invention. The functions specifically implemented by the computer-readable storage medium embodiment of the present invention are the same as those of the above method embodiment, and the beneficial effects achieved are also the same as those of the above method.

[0103] Embodiments of the present invention also disclose a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the foregoing method.

[0104] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order mentioned in the operational illustrations. For example, depending on the functions / operations involved, two consecutive blocks shown may actually be executed substantially simultaneously or the blocks can sometimes be executed in the reverse order. In addition, the embodiments presented and described in the flowcharts of the present invention are provided by way of example for the purpose of providing a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated, in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0105] Furthermore, although the present invention has been described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features described may be integrated in a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It should also be understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present invention. Rather, given the attributes, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the modules will be understood within the ordinary skill of an engineer. Thus, those skilled in the art can implement the present invention as set forth in the claims without undue experimentation. It should also be understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0106] If the above-mentioned functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs, Read-Only Memories), random access memories (RAMs, Random Access Memories), magnetic disks, or optical discs.

[0107] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a predefined sequence of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution device, apparatus, or equipment (such as a computer-based device, a device including a processor, or other devices that can fetch instructions from the instruction execution device, apparatus, or equipment and execute the instructions), or in combination with these instruction execution devices, apparatuses, or equipment. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by or in combination with an instruction execution device, apparatus, or equipment.

[0108] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or, if necessary, other suitable processing, and then storing it in a computer memory.

[0109] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0110] In the description of this specification, the descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0111] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the claims and their equivalents.

[0112] The above has specifically described the preferred embodiments of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art can also make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of the present invention.

Claims

1. An OSAHS detection method, characterized in that, it includes: Obtain the original blood oxygen signal data, perform data preprocessing on the original blood oxygen signal data to obtain blood oxygen signal slice data; Use a one-dimensional neural network classifier to analyze the blood oxygen signal slice data to obtain a first-stage classification result; wherein, the one-dimensional neural network classifier is trained and generated by the blood oxygen signal data with labeled sleep apnea events; the first-stage classification result includes normal breathing, suspected hypoventilation, and apnea; Use a classifier based on rule matching to analyze the blood oxygen signal slice data of the target segment to obtain a second-stage classification result; wherein, the classifier based on rule matching is generated by using the scores of blood oxygen signal statistical indicators as the discrimination criterion; the target segment includes the segments corresponding to the blood oxygen signal slice data with the first-stage classification results of normal breathing and suspected hypoventilation; the second-stage classification result includes normal breathing, hypoventilation, and apnea; wherein, the method further includes: Set the discrimination rule of the scores of the blood oxygen signal statistical indicators based on blood oxygen saturation, blood oxygen signal variance, and blood oxygen minimum value; Create a classifier based on rule matching according to the discrimination rule.

2. The OSAHS detection method according to claim 1, characterized in that, The obtaining of the original blood oxygen signal data, performing data preprocessing on the original blood oxygen signal data to obtain blood oxygen signal slice data includes: Obtain the original blood oxygen signal data of the target object in a preset time period; Based on a preset proportion threshold of blood oxygen artifacts, screen the original blood oxygen signal data to obtain target blood oxygen signal data; Based on a preset duration, perform data slicing on the target blood oxygen signal data; Perform downsampling and normalization on the target blood oxygen signal data after data slicing to obtain blood oxygen signal slice data.

3. The OSAHS detection method according to claim 1, characterized in that, it further includes: Create a one-dimensional neural network classifier based on the blood oxygen signal data with labeled sleep apnea events.

4. The OSAHS detection method according to claim 3, characterized in that, The creating of the one-dimensional neural network classifier based on the blood oxygen signal data with labeled sleep apnea events includes: Determine the training samples according to the blood oxygen signal data with labeled sleep apnea events; Based on a one-dimensional convolutional operator, a one-dimensional max pooling operator, an activation function, and a fully connected layer, set a neural network model with two branches, perform classification training on the neural network model based on the training samples, and adjust the neural network model based on the training results to obtain a one-dimensional neural network classifier; wherein, both branches of the neural network model include a number of the one-dimensional convolutional operators, the one-dimensional max pooling operators, and the activation functions; the two branches of the neural network model are connected at the fully connected layer.

5. The OSAHS detection method according to claim 1, characterized in that, The one-dimensional neural network classifier includes two branches. By using the one-dimensional neural network classifier to analyze the sliced blood oxygen signal data, a first-stage classification result is obtained, including: According to each segment in the sliced blood oxygen signal data, segment extension is performed to obtain the input of the first branch; and segment combination is performed to obtain the input of the second branch; Convolution operations, activation responses, and dimensionality reduction are respectively performed on the input of the first branch and the input of the second branch to correspondingly obtain a first feature map and a second feature map; The first feature map and the second feature map are concatenated according to the channel dimension to obtain a third feature map; Full connection classification is performed on the third feature map to obtain a first-stage classification result.

6. According to the OSAHS detection method described in claim 1, characterized in that By using a classifier based on rule matching to analyze the sliced blood oxygen signal data of the target segment, a second-stage classification result is obtained, including: According to the descending amplitude of the blood oxygen saturation of the sliced blood oxygen signal data of the target segment, a first scoring process is performed; According to the fluctuation amplitude of the variance of the blood oxygen signal of the sliced blood oxygen signal data of the target segment, a second scoring process is performed; According to the descending amplitude of the minimum blood oxygen value of the sliced blood oxygen signal data of the target segment, a third scoring process is performed; Based on the first scoring process, the second scoring process, and the third scoring process, the score of the blood oxygen signal statistical index is determined; According to the score of the blood oxygen signal statistical index, a second-stage classification result is obtained.

7. An OSAHS detection system, characterized in that including: A first module for acquiring original blood oxygen signal data and performing data preprocessing on the original blood oxygen signal data to obtain sliced blood oxygen signal data; A second module for using a one-dimensional neural network classifier to analyze the sliced blood oxygen signal data to obtain a first-stage classification result; wherein, the one-dimensional neural network classifier is trained and generated by the blood oxygen signal data with labeled sleep apnea events; the first-stage classification result includes normal breathing, suspected hypopnea, and apnea; A third module for using a classifier based on rule matching to analyze the sliced blood oxygen signal data of the target segment to obtain a second-stage classification result; wherein, the classifier based on rule matching is generated by using the score of the blood oxygen signal statistical index as a discrimination criterion; the target segment includes the segments corresponding to the sliced blood oxygen signal data with the first-stage classification results of normal breathing and suspected hypopnea; the second-stage classification result includes normal breathing, hypopnea, and apnea; wherein, the system further includes a module for performing the following operations: Setting the discrimination rule for the score of the blood oxygen signal statistical index based on blood oxygen saturation, variance of the blood oxygen signal, and minimum blood oxygen value; According to the discrimination rule, creating a classifier based on rule matching.

8. An OSAHS detection device includes a processor and a memory; The memory is used for storing programs; The processor executes the program to implement the method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The storage medium stores a program, and the program is executed by a processor to implement the method according to any one of claims 1 to 6.

Images