CN-122025187-A - CBT-i-based omniseal record sleep supervision system

CN122025187ACN 122025187 ACN122025187 ACN 122025187ACN-122025187-A

Abstract

The invention belongs to the technical field of digital monitoring, provides a CBT-i-based omnism recording sleep supervision system, and aims to solve the problems that an insomnia patient is difficult to accurately record the occurrence time of omnism in sleep, and the intervention effect of CBT-i is affected; the method comprises the steps of collecting heart rate and respiratory data through an intelligent bracelet by a data collecting module, predicting a variational high-emission window through an LSTM+attention model after preprocessing, screening the variational high-emission window twice according to historical variational data by a screening module to obtain a reference variational high-emission window set, pushing a new variational timestamp range by a new variational analysis module through a time matching method, building a suspected set through time overlapping matching, calculating matching degree by a time distribution determining module, selecting an optimal window as a new variational accurate timestamp range, and providing accurate data support for CBT-i.

Inventors

ZHANG JUNHANG
LI ZHIJIAN
Song Mingfen
YU ZHENGHE
XU QIANQIAN
XU YOU
HUANG YUCHEN
YANG LILI
REN LISHAN
Wei Youdan

Assignees

杭州市第七人民医院(杭州市心理危机研究与干预中心)
杭州心蜗科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260407

Claims (10)

1. The CBT-i based omniseal record sleep supervision system is characterized by comprising: the data acquisition module acquires real-time physiological data of a user, analyzes the real-time physiological data of the user by utilizing a variegated high-frequency window prediction model, and predicts to obtain a plurality of discontinuous variegated high-frequency windows; The variegated high-frequency window screening module is used for carrying out primary screening on a plurality of discontinuous variegated high-frequency windows according to the historical variegated data of the user to obtain a reference variegated high-frequency window set, and carrying out secondary screening on the reference variegated high-frequency window set by combining the real-time physiological data of the user and the historical variegated data of the user to obtain a reference variegated high-frequency window set; The new omnirange data analysis module adopts a time matching method and combines the historical omnirange data of the user to infer the timestamp distribution range of the newly recorded user omnirange data; performing matching analysis on the timestamp distribution range of the newly recorded user variegated data and the reference variegated high-frequency window set by using a time overlapping method, if so, reserving the matched reference variegated high-frequency window, and establishing a suspected variegated high-frequency window set; The new omnirange data time distribution determining module is used for calculating the matching degree of the newly recorded user omnirange data in any one of the suspected omnirange high-frequency windows in the suspected omnirange high-frequency window set based on the established suspected omnirange high-frequency window set and combining with the user history omnirange data, and selecting the starting time and the ending time corresponding to the suspected omnirange high-frequency window with the largest matching degree as the time stamp distribution range of the newly recorded user omnirange data.
2. The CBT-i based omniseal logging sleep management system of claim 1, wherein the user real-time physiological data comprises: collecting heart rate data, namely collecting continuous heart rate data through a PPG optical heart rate sensor of the intelligent bracelet; Collecting respiratory data, namely analyzing respiratory characteristics through tiny acceleration changes of respiratory motion of the chest/abdomen by adopting an intelligent bracelet internal acceleration sensor; The real-time physiological data of the user specifically comprises heart rate frequency data and respiratory frequency data.
3. The CBT-i based omnisnography sleep monitoring system of claim 1, wherein the predicting comprises the following steps: Removing motion artifact of heart rate data, namely judging that the heart rate data has the motion artifact when the amplitude of an acceleration sensor is more than 0.5g by combining with intelligent bracelet acceleration sensor data; repairing motion artifact by adopting an adjacent 5-second linear interpolation method, extracting time domain features and frequency domain features by taking 5 minutes as a window, and carrying out standardization processing on the time domain features and the frequency domain features by adopting a Z-score standardization method; carrying out breathing characteristic extraction and denoising treatment on breathing data, denoising the breathing data through bandpass filtering with the preset size of 0.1-0.5 Hz by using intelligent bracelet acceleration sensor data, and identifying wave peaks and wave troughs through peak detection; Abnormal value correction, namely if the inspiration/expiration ratio Ti/Te exceeds the range of 0.8-1.2, judging that the abnormal breathing mode is an abnormal breathing mode, and adopting an adjacent 5-second linear interpolation method for correction; The preprocessed breathing data and heart rate data are used for an input layer of the large model; The long-term memory network LSTM is used for capturing long time sequence dependence of heart rate and respiration, and the Attention mechanism Attention is used for focusing on key characteristic periods: The variational high-altitude window prediction model comprises 1 input layer, 1 LSTM layer, 1 Attention layer, 1 full-connection output layer and a plurality of discontinuous variational high-altitude windows, wherein the input layer is used for inputting preprocessed respiratory data and heart rate data, the LSTM layer comprises 128 hidden units and is used for outputting hidden states of all time steps, the Attention layer is used for calculating Attention weights of all time steps of LSTM by adopting additive Attention and obtaining global context vectors by weighting and summing, the full-connection output layer comprises 64 full-connection layers of ReLU activated neurons, and finally the discontinuous variational high-altitude windows are obtained by outputting single neurons activated by Sigmoid.
4. The system for monitoring and managing the sleep of the omnisnography based on the CBT-i is characterized in that the concrete process of primarily screening a plurality of discontinuous omnisnography windows to obtain a reference omnisnography window set is as follows: The user history omnirange data specifically comprises two types of core information, namely, a first type of time stamp of omnirange generation time in the past of a user and used for marking the accurate time of omnirange generation, and a second type of user real-time physiological data which corresponds to the time stamp one by one, namely heart rate data and respiratory data physiological index data which are synchronously collected at the omnirange generation time; through the historical omniseal data of the user, a plurality of discontinuous omniseal high-incidence windows can be initially screened, and the specific screening process is as follows: Aiming at a plurality of discontinuous variegated high-frequency windows predicted by a variegated high-frequency window prediction model, acquiring the starting time and the ending time of each variegated high-frequency window, and subtracting the starting time of the variegated high-frequency window from the ending time of the variegated high-frequency window to acquire each variegated high-frequency window interval.
5. The system for CBT-i based omnisnography sleep supervision of claim 4, wherein the specific process of initially screening a plurality of discrete omnisnography windows to obtain a set of reference omnisnography windows further comprises: presetting a time window, marking as a coverage time window, avoiding missed detection caused by small fluctuation of the occurrence period of the user's omnism, and expanding the search range of the user's history omnism data on the basis of a plurality of discontinuous omnism high-frequency windows obtained by prediction: The method comprises the following steps of obtaining the variegated high-frequency window intervals of all the variegated high-frequency windows, and carrying out the following operations on any one variegated high-frequency window interval: Forward expanding a variegated high-frequency window interval, namely, for any one variegated high-frequency window interval, forward extending a coverage time window by taking the starting time of the variegated high-frequency window interval as a reference; the method comprises the steps of expanding a variegated high-frequency window interval backwards, namely, for any one variegated high-frequency window interval, extending a coverage time window forwards by taking the ending time of the variegated high-frequency window interval as a reference; The high-probability window interval after forward and backward expansion is recorded as an expansion interval; The time length setting of the coverage time window needs to satisfy a non-coincidence principle, namely after all discontinuous variegated high-emission window intervals are respectively expanded forwards and backwards, no time overlapping part exists between any two adjacent expansion intervals, and repeated counting is avoided when the historical time stamps are counted later; the method comprises the steps of calling historical idea data of a user within 90 days, and counting the number of time stamps of the occurrence of the idea in any expansion interval; If the number of the time stamps of any one expansion interval exceeds 30% of the total number in the user history miscellaneous data, reserving the miscellaneous high-frequency window corresponding to the expansion interval with the number of the time stamps exceeding 30% of the total number; Reserving the variegated high-frequency windows corresponding to the expansion intervals with the proportion exceeding 30%, and establishing a reference variegated high-frequency window set according to the time sequence of the variegated high-frequency windows.
6. The CBT-i based omnisnography sleep monitoring system of claim 1, wherein the secondary screening of the reference omnisnography window set comprises the following steps: Extracting user real-time physiological data corresponding to each user history miscellaneous record from the user history miscellaneous data; Acquiring real-time physiological data of a user acquired in an expansion interval corresponding to each reference variational high-frequency window, and carrying out average processing on heart rate frequency data and respiratory frequency data to acquire a real-time heart rate frequency average value and a real-time respiratory frequency average value; And carrying out average processing on the heart rate data and the respiratory frequency data to obtain a historical heart rate frequency average value and a historical respiratory frequency average value.
7. The system for CBT-i based omnisnography sleep supervision system of claim 6, wherein the secondary screening of the set of reference omnisnography windows further comprises: the method comprises the following specific processes of performing deviation comparison on a real-time heart rate frequency average value and a real-time respiratory frequency average value and a historical heart rate frequency average value and a historical respiratory frequency average value: If the deviation between the real-time heart rate frequency mean value and the historical heart rate frequency mean value is more than 10%, judging that the heart rate frequency mean value and the historical heart rate frequency mean value are inconsistent; if the deviation between the real-time respiratory frequency mean value and the historical respiratory frequency mean value is more than 10%, judging that the respiratory frequency mean value and the historical respiratory frequency mean value are inconsistent; if the deviation between the real-time heart rate frequency mean value and the historical heart rate frequency mean value is less than 10 percent and the deviation between the real-time respiratory frequency mean value and the historical respiratory frequency mean value is less than 10 percent, judging to be consistent; and reserving the standard variegated high-frequency windows which are judged to be consistent, and establishing and obtaining a reference variegated high-frequency window set according to the time sequence of the standard variegated high-frequency windows.
8. The CBT-i based omniseal logging sleep supervision system according to claim 1, wherein the specific process of predicting the timestamp distribution range of the newly logged user omniseal data is as follows: the newly recorded user omnism data is specifically that the user cannot determine the accurate time when omnism occurs, and only the approximate range of the occurrence of omnism can be determined, including but not limited to pre-sleep omnism or post-sleep omnism; calculating the average sleep time of the user in the past 90 days as a standard for distinguishing the pre-sleep omnirange from the post-sleep omnirange; Obtaining time stamps in the user history miscellaneous data in the past 90 days, dividing 1 day into 24 units according to one hour as one time unit, numbering the time units according to a time sequence by taking 00:00 as the start, counting the number of the time stamps in the user history miscellaneous data in the past 90 days in each time unit, and calculating the duty ratio of each time unit; firstly marking time units with the proportion of more than or equal to 10% as high-frequency units, and identifying time continuous unit groups from the high-frequency units; Calculating the cumulative duty ratio of each continuous unit group, and preferentially selecting the largest continuous unit group with the cumulative duty ratio more than or equal to 80 percent, if the single group is less than 80 percent, combining the adjacent sub-high frequency continuous unit groups until the cumulative duty ratio is more than or equal to 80 percent; Integrating the periods corresponding to the screened continuous unit groups to form an initial timestamp distribution range; taking average falling time as a boundary, splitting an initial timestamp distribution range into a pre-sleep sub-range and a post-sleep sub-range, wherein the specific splitting process is as follows: A pre-sleep sub-range, which is a portion of the initial range that belongs to a pre-sleep period; a post-sleep sub-range, which is a portion of the initial range that belongs to a post-sleep period; Selecting a corresponding sub-range as a final timestamp distribution range according to pre-sleep miscellaneous/post-sleep miscellaneous in newly recorded user miscellaneous data: if the newly recorded user miscellaneous data is the pre-sleep miscellaneous, outputting a pre-sleep sub-range; And outputting a sub-range after sleep if the newly recorded user miscellaneous data is the post-sleep miscellaneous.
9. The CBT-i based omnisnography sleep monitoring system of claim 1, wherein the concrete process of establishing the suspected omnisnography high-frequency window set is as follows: Acquiring a variational high-frequency window interval of each reference variational high-frequency window in the reference variational high-frequency window set; Based on the pre-sleep sub-range and the post-sleep sub-range, performing time overlapping judgment on the omnism high-frequency window interval of each reference omnism high-frequency window in the reference omnism high-frequency window set and the pre-sleep sub-range and the post-sleep sub-range; if the starting time of the omniscent high-frequency window interval of the reference omniscent high-frequency window is smaller than or equal to the ending time of the sub-range before sleeping and the ending time of the omniscent high-frequency window interval of the reference omniscent high-frequency window is larger than or equal to the starting time of the sub-range before sleeping; If the starting time of the omniscent high-frequency window interval of the reference omniscent high-frequency window is smaller than or equal to the ending time of the sub-range after sleeping and the ending time of the omniscent high-frequency window interval of the reference omniscent high-frequency window is larger than or equal to the starting time of the sub-range after sleeping; and collecting all the reference variegated high-frequency windows which are judged to be overlapped, and establishing a suspected variegated high-frequency window set.
10. The CBT-i based omniseal record sleep supervision system according to claim 1, wherein the specific process of selecting the start time and the end time corresponding to the suspected omniseal high-altitude window with the largest matching degree as the timestamp distribution range of the newly recorded user omniseal data is as follows: Acquiring real-time physiological data corresponding to the final timestamp distribution range of newly recorded user variational data, calculating heart rate frequency mean value and marking as And respiratory rate mean value is recorded as ; Acquiring a historical heart rate frequency mean value corresponding to any one suspected variational high frequency window in the suspected variational high frequency window set and marking the historical heart rate frequency mean value as And historical respiratory rate mean value is recorded as ; Calculating the matching degree of the newly recorded user variegated data in any suspected variegated high-frequency window in the suspected variegated high-frequency window set by using Euclidean distance formula ; And selecting the starting time and the ending time corresponding to the suspected variegated high-altitude window with the largest matching degree as the timestamp distribution range of the newly recorded user variegated data.

Description

CBT-i-based omniseal record sleep supervision system Technical Field The invention belongs to the technical field of sleep health medical treatment, and particularly relates to a CBT-i-based omniscent recording sleep supervision system. Background The insomnia cognitive behavior therapy (CBT-i) is an insomnia intervention means widely applied in clinic, one of the key implementation links is to record negative cognition (namely 'omnirange') related to sleep, such as monitoring and recording of negative cognition of sleep anxiety, ruminant thinking of the next day transaction and the like, so as to provide basis for subsequent analysis of sleep disorder causes and establishment of an intervention scheme. However, in a practical application scenario, it is often difficult for an insomnia patient to accurately record an accurate time point of occurrence of a omnism during sleep. In particular, when the omnism is generated, most of the insomnia patients usually make up the approximate time of the occurrence of the omnism after waking up, rather than accurately recording in real time, because of the consideration of factors such as not affecting the integrity and continuity of sleep. The non-real-time and fuzzy recording mode directly causes the problems of incomplete sleep data and insufficient integrity, namely, due to the lack of an accurate timestamp distribution range, the CBT-i therapy is difficult to realize accuracy when analyzing the association mechanism of 'omnism' and sleep disorder and formulating a personalized intervention scheme, so that the intervention effect is adversely affected. Therefore, how to accurately record the distribution range of the timestamps of the occurrence time of the omnism in the sleeping process of the insomnia patient and provide complete and accurate sleeping data support for the CBT-i therapy becomes a technical problem to be solved in the field. To this end, the invention provides a CBT-i based omnisnoge sleep supervision system. Disclosure of Invention In order to overcome the deficiencies of the prior art, at least one technical problem presented in the background art is solved. The technical scheme adopted for solving the technical problems is as follows: one of the purposes of the present invention is to provide a CBT-i based omnisnography sleep supervision system, comprising: the data acquisition module acquires real-time physiological data of a user, analyzes the real-time physiological data of the user by utilizing a variegated high-frequency window prediction model, and predicts to obtain a plurality of discontinuous variegated high-frequency windows; The variegated high-frequency window screening module is used for carrying out primary screening on a plurality of discontinuous variegated high-frequency windows according to the historical variegated data of the user to obtain a reference variegated high-frequency window set, and carrying out secondary screening on the reference variegated high-frequency window set by combining the real-time physiological data of the user and the historical variegated data of the user to obtain a reference variegated high-frequency window set; The new omnirange data analysis module adopts a time matching method and combines the historical omnirange data of the user to infer the timestamp distribution range of the newly recorded user omnirange data; performing matching analysis on the timestamp distribution range of the newly recorded user variegated data and the reference variegated high-frequency window set by using a time overlapping method, if so, reserving the matched reference variegated high-frequency window, and establishing a suspected variegated high-frequency window set; The new omnirange data time distribution determining module is used for calculating the matching degree of the newly recorded user omnirange data in any one of the suspected omnirange high-frequency windows in the suspected omnirange high-frequency window set based on the established suspected omnirange high-frequency window set and combining with the user history omnirange data, and selecting the starting time and the ending time corresponding to the suspected omnirange high-frequency window with the largest matching degree as the time stamp distribution range of the newly recorded user omnirange data. As a further improvement of the present invention, the user real-time physiological data specifically includes: collecting heart rate data, namely collecting continuous heart rate data through a PPG optical heart rate sensor of the intelligent bracelet; Collecting respiratory data, namely analyzing respiratory characteristics through tiny acceleration changes of respiratory motion of the chest/abdomen by adopting an intelligent bracelet internal acceleration sensor; The real-time physiological data of the user specifically comprises heart rate frequency data and respiratory frequency data. As a further improvement of the present invention, the specific