An apparatus for analyzing depression level based on machine learning algorithm and physiological data
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING XUEYANG TECH CO LTD
- Filing Date
- 2023-08-15
- Publication Date
- 2026-07-21
AI Technical Summary
In existing technologies for analyzing depression levels based on machine learning, the collection of physiological data is prone to noise, leading to low accuracy of the analysis results.
The system employs a primary validation module and a secondary validation module based on machine learning algorithms to perform self-validation and inter-class validation on various types of user physiological data. Through noise reduction processing, the accuracy of the physiological data is ensured, thereby improving the accuracy of depression level analysis.
Accurate denoising of physiological data was achieved, ensuring that the input data of the depression severity analysis model can accurately represent the user's normal physiological data, thus improving the accuracy of depression severity analysis.
Smart Images
Figure CN117198539B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of psychological assessment technology, and in particular to a device for analyzing the degree of depression based on machine learning algorithms and physiological data. Background Technology
[0002] Currently, with the increasing intelligence of smart devices and the widespread public attention to mental health, many smart devices or software with functions to analyze depression levels or assess mental health have emerged on the market. The implementation of such technologies allows users to monitor and assess their mental health without having to seek medical treatment in person, making the psychological assessment process simpler and more intelligent.
[0003] However, existing intelligent depression severity analysis technologies mostly rely on machine learning to establish a depression severity analysis system or model from numerous cases. Based on this system or model, they analyze various physiological data of users. However, due to the diversity of physiological data types and the varying psychological and physiological states of different individuals within the cases, the collected physiological data is prone to noise. For example, EEG parameters, a commonly used decision variable for analyzing depression severity, exhibit significant differences depending on the user's movement and physical health status. Therefore, simply using the collected physiological data as the depression severity analysis system or model in existing technologies will lead to substantial errors in the output depression severity analysis results. For example, as illustrated in publication number CN11621... Chinese invention patent 1322A, published on June 6, 2023, entitled "A Method and System for Depression Recognition Based on Machine Learning EEG Signals," discloses a method and system for depression recognition based on machine learning EEG signals. The method includes: acquiring raw EEG signals; denoising and normalizing the raw EEG information to obtain a first EEG signal; extracting features from the first EEG signal, including linear features, nonlinear features, and power spectrum features; inputting the extracted linear features, nonlinear features, and power spectrum features into a random forest for training to obtain a trained depression classification model; and outputting the depression probability of the target EEG signal based on the trained depression classification model. The method and system pre-select features from the acquired EEG signals, choosing linear features, nonlinear features, and power spectrum features, and after normalizing the extracted features, select different EEG channels to construct a random forest model. However, this patent does not denoise the acquired physiological data, i.e., the EEG signals, leading to unreliable accuracy in the subsequently obtained depression probability.
[0004] Therefore, this invention proposes a device for analyzing the degree of depression based on machine learning algorithms and physiological data. Summary of the Invention
[0005] This invention provides a device for analyzing depression levels based on machine learning algorithms and physiological data. It performs self-validation and inter-class validation on all types of physiological data of the user to achieve accurate denoising of the collected physiological data. This ensures that the final denoised data input into the depression level analysis model can represent the user's normal physiological data and makes the analyzed depression level more accurate.
[0006] This invention provides a device for analyzing the degree of depression based on machine learning algorithms and physiological data, comprising:
[0007] The initial verification module is used to acquire multiple types of physiological data from users in real time, perform self-verification on each type of physiological data, and obtain denoised data to be judged for each type of physiological data.
[0008] The secondary verification module is used to perform inter-class verification denoising on all types of physiological data to be judged, and obtain the final denoised physiological data.
[0009] The model building module is used to build a depression severity analysis model based on machine learning algorithms.
[0010] The severity determination module is used to input the temporal fluctuation characteristics of the final denoised physiological data into the depression severity analysis model to obtain the user's current depression severity value.
[0011] The preferred initial verification module includes:
[0012] The outlier denoising submodule is used to generate multiple single-category physiological data recording curves for users based on real-time acquired physiological data of each category, and to remove outliers in the single-category physiological data recording curves to obtain single-category non-outlier recording curves.
[0013] The curve representation submodule is used to generate a physiological data change representation function that represents the physiological data changes over time in a single non-outlier recording curve.
[0014] The hierarchy determination submodule is used to determine the hierarchy required for trend analysis of physiological data of each type and the trend variability tolerance of physiological data of the current type at different trend analysis hierarchy.
[0015] The data denoising submodule is used to determine the denoised data of the current type of physiological data in a single non-outlier recording curve based on the physiological data change representation function of each type of physiological data, the corresponding trend analysis level required, and the trend gradient tolerance at different trend analysis levels.
[0016] Preferably, the hierarchical determination submodule includes:
[0017] The data representation unit is used to generate a physiological data change representation function for each category of each computational sample based on the physiological data of users in a large number of computational samples that have been de-noised from outliers.
[0018] The tolerance determination unit is used to determine the range of physiological data differences between adjacent time points in each physiological data of the same category of all calculated samples as the trend gradient tolerance of the first trend analysis level.
[0019] The range determination unit is used to calculate the first derivative value of the physiological data change representation function at each time step, and determine the range of all first derivative values of the physiological data of each calculation sample as the range of values to be judged.
[0020] The hierarchy determination unit is used to determine the required hierarchy for trend analysis of physiological data of each type and the trend gradient tolerance of physiological data of the current type at different trend analysis levels, based on the range of values to be judged for physiological data of all calculated samples.
[0021] Preferably, the hierarchical determination unit includes:
[0022] The first determining subunit is used to determine the range of differences between all adjacent second derivative values of the physiological data of each calculated sample when the commonality of the range of values to be judged for the same type of physiological data of all calculated samples is not less than the commonality threshold. This range is then used as the new range of values to be judged until the commonality of the range of values to be judged for the current type of physiological data of all calculated samples is less than the commonality threshold. Then, the value of the number of judgments executed in the previous execution is used as the value of the level required for trend analysis of the current type of physiological data, and the range of values to be judged in each judgment process executed in the previous and previous executions is used as the trend gradient tolerance value of the current type of physiological data at the corresponding trend analysis level.
[0023] The second determining subunit is used to treat the value of the number of judgments executed in the previous execution as the value of the level required for trend analysis of the current type of physiological data when the commonality of the range of values to be judged for the same type of physiological data in all calculated samples is less than the commonality threshold, and to treat the range of values to be judged in the previous execution as the trend gradient tolerance value of the current type of physiological data at the corresponding trend analysis level.
[0024] Preferably, the data denoising submodule includes:
[0025] The corresponding derivative unit is used to perform corresponding derivatives on the corresponding physiological data change representation function based on the required level of trend analysis for each type of physiological data, to obtain multiple derivative functions of different orders, and to determine the different derivative values of multiple derivative functions of different orders at each time step.
[0026] The data denoising unit is used to take the physiological values of different order derivatives of the trend gradient tolerance value that exceed the corresponding trend analysis level in the corresponding single non-outlier record curve as the denoised data to be judged for the current type of physiological data.
[0027] Preferably, the secondary verification module includes:
[0028] The temporal determination submodule is used to determine the temporal interval of all data to be denoised in the corresponding category of physiological data;
[0029] The commonality calculation submodule is used to calculate the number of times each time point exists in all types of denoised data to be judged in the time interval of all types of data. Based on the number of existences and the total number of types of physiological data, the inter-class commonality of each time point is determined.
[0030] The inter-class verification submodule is used to perform inter-class verification and denoising on the physiological data to be judged for all types of data based on inter-class commonality, so as to obtain the final denoised physiological data.
[0031] Preferably, the inter-class validation submodule includes:
[0032] The time determination unit is used to treat the time when the inter-class commonality is less than the inter-class commonality threshold as the retained time, and treat the remaining time in all time intervals other than the retained time as the time to be denoised.
[0033] The final denoising unit is used to remove the physiological values at the time points that need to be denoised from all types of physiological data to obtain the final denoised physiological data.
[0034] Preferably, the model building module includes:
[0035] The instance acquisition submodule is used to acquire a preset amount of denoised physiological data of multiple types and corresponding professional assessment values of depression level for assessment instances.
[0036] The sample extraction submodule is used to treat the temporal fluctuation characteristics of all denoised physiological data in each evaluation instance as a single sample.
[0037] The model building submodule is used to divide all samples into training samples and test samples, and to train and test based on the training samples and test samples to obtain a depression level analysis model.
[0038] Among them, temporal fluctuation characteristics include, but are not limited to, extreme values of individual types of physiological data and time-varying functional representations.
[0039] Preferred options also include:
[0040] The early warning generation module is used to generate early warning information based on the user's current depression level.
[0041] The early warning information display module is used to display early warning information on one end or multiple ends.
[0042] Preferably, the early warning information display module includes:
[0043] The instruction receiving submodule is used to receive the user's display permission setting instructions and parse the display mode from the display permission restriction instructions.
[0044] The first display submodule is used to display the warning information on the corresponding single end when the display mode is single-end display;
[0045] The second display submodule is used to determine the target multi-devices corresponding to the display permissions when the display mode is multi-device display, and to display the warning information to the corresponding target multi-devices.
[0046] The beneficial effects of this invention compared to existing technologies are as follows: based on self-verification and inter-class verification of all types of physiological data of users, accurate denoising of the collected physiological data is achieved, thereby ensuring that the final denoised data input into the depression level analysis model can represent the user's normal physiological data and make the analyzed depression level more accurate.
[0047] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.
[0048] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0049] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0050] Figure 1 This is a schematic diagram of a device for analyzing the degree of depression based on machine learning algorithms and physiological data, as described in an embodiment of the present invention.
[0051] Figure 2 This is a schematic diagram of the initial verification module in an embodiment of the present invention;
[0052] Figure 3 This is a schematic diagram of the hierarchy determination submodule in an embodiment of the present invention. Detailed Implementation
[0053] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0054] Example 1:
[0055] This invention provides a device for analyzing the degree of depression based on machine learning algorithms and physiological data, referencing... Figure 1 ,include:
[0056] The initial verification module is used to acquire multiple types of physiological data from users in real time (such as pulse wave signals, electroencephalogram signals, skin conductance signals, gastric conductance signals, electromyography signals, electrooculogram signals, etc.). It performs self-verification on each type of physiological data (that is, based on the trend changes and values of the physiological data of that type, it verifies and filters the physiological data of that type to filter out suspected noisy data that needs further inter-class verification). It obtains the denoised data to be judged for each type of physiological data (that is, the part of the physiological data of a single type that does not meet the corresponding conditions during the self-verification process). Through self-verification, it realizes the verification of its own data changes based on the data change patterns of the physiological data itself.
[0057] The secondary verification module is used to perform inter-class verification denoising on the physiological data to be judged for all types of physiological data (that is, to realize the second verification and screening of the physiological data to be judged based on the degree of commonality of the physiological data to be judged in time series, and to filter out the final denoised physiological data). The final denoised physiological data is obtained (that is, the remaining physiological data after deleting the part of the physiological data that was judged as noise data after self-verification and inter-class verification). Based on the inter-class verification process, the physiological data to be judged suspected as noise data identified in the previous self-verification is further verified in time series, avoiding the occurrence of over-denoising. That is, the physiological data is accurately denoised, ensuring that the final denoised data input into the depression level analysis model can represent the user's normal physiological data.
[0058] The model building module is used to build a depression level analysis model based on machine learning algorithms (the temporal fluctuation characteristics of the final denoised physiological data obtained after self-validation and inter-class validation from the physiological data of multiple users are divided into training samples and test samples, and then input into random forest for training and testing to build a depression level analysis model).
[0059] The degree determination module is used to input the temporal fluctuation characteristics of the final denoised physiological data (e.g., the first derivative value of the first derivative function representing the change of the final physiological data over time at each moment, or the maximum and minimum values, mean, etc. of the final denoised physiological data in a single analysis period) into the depression degree analysis model to obtain the user's current depression degree value (i.e., the value representing the user's current depression degree value).
[0060] By using the temporal fluctuation characteristics of physiological data after multiple denoising processes as input to the depression severity analysis model, accurate depression severity analysis results can be obtained.
[0061] Example 2:
[0062] Based on Example 1, the initial verification module, refer to Figure 2 ,include:
[0063] The outlier denoising submodule is used to generate multiple single-category physiological data recording curves (i.e., curves recording the changes of physiological data of a single category over time) for users based on real-time acquired physiological data of each category. Outliers in the single-category physiological data recording curves are deleted (outliers are, for example, numerical points that cannot be connected to physiological values at adjacent time points, or numerical points in the curve where the value changes abruptly, where the physiological value is a single value in the single-category physiological data recording curve), and a single non-outlier recording curve is obtained (i.e., the single-category physiological data recording curve after deleting outliers). Outliers in each category of physiological data have been deleted.
[0064] The curve characterization submodule is used to generate a physiological data change characterization function (i.e., a function used to represent the physiological values in the physiological data of a single non-outlier recording curve as a function of time) to characterize the physiological data changes over time in a single non-outlier recording curve; it realizes the quantification of the relationship between physiological data and time changes.
[0065] The hierarchy determination submodule is used to determine the hierarchy required for trend analysis of each type of physiological data (i.e., the number and order of derivatives when performing trend analysis on a single type of physiological data; the hierarchy required for trend analysis includes: when the hierarchy required for trend analysis is 1, the numerical range of the single non-outlier recording curve of the current type of physiological data is directly analyzed, and the numerical range of the first derivative of the characteristic function of the change of the current type of physiological data at each time step is analyzed; when the hierarchy required for trend analysis is 2, the numerical range of the second derivative of the characteristic function of the change of the current type of physiological data at each time step is also analyzed, and so on) and the trend tolerance of the current type of physiological data at different hierarchy of trend analysis. That is, the numerical range of a single non-outlier recording curve of the current type of physiological data. The trend gradient tolerance includes: when the value of the level required for trend analysis is 1, the numerical range of the first derivative of the physiological data change representation function of the current type at each time step; when the value of the level required for trend analysis is 2, the numerical range of the second derivative of the physiological data change representation function of the current type at each time step, and so on. This enables the classification and analysis of the trend change rules of physiological data, and uses the level required for trend analysis and the trend gradient tolerance corresponding to different trend analysis levels to differentiate the trend change rules of different types of physiological data, thus realizing a reasonable representation of the conventional trend change rules of different types of physiological data.
[0066] The data denoising submodule is used to determine the denoised data of the current type of physiological data in a single non-outlier recording curve based on the physiological data change representation function of each type of physiological data, the corresponding trend analysis level, and the trend gradient tolerance at different trend analysis levels. Based on the aforementioned determination of the conventional trend change rules of different types of physiological data, it achieves the screening of physiological values that do not conform to the conventional trend change rules of the corresponding type, thereby achieving the preliminary accurate screening of suspected noise data.
[0067] Example 3:
[0068] Based on Example 2, the sub-modules are determined hierarchically, with reference to... Figure 3 ,include:
[0069] The data representation unit is used to generate a physiological data change representation function for each type of each computational sample (i.e., a function representation of the physiological values in the single non-outlier record curve contained in the computational sample that have been de-noised from outliers) based on the physiological data of users in a large number of computational samples (i.e., samples containing a single non-outlier record curve obtained after removing outliers from the physiological data of other users).
[0070] The tolerance determination unit is used as the range of values of the physiological data difference between adjacent time points in each physiological data of the same type of all calculated samples (i.e., the range of values formed by the difference of physiological values between adjacent time points in all physiological data of the same type of all calculated samples) as the trend gradient tolerance of the first trend analysis level (i.e., the analysis level that directly performs trend analysis on a single non-outlier record curve).
[0071] The range determination unit is used to calculate the first derivative value of the physiological data change representation function at each time step, and determine the range of all first derivative values of the physiological data of each calculation sample as the range of values to be judged.
[0072] The above process enables multi-level trend analysis of different types of single-type non-outlier recording curves by determining the range of values for the single-type non-outlier recording curve at each time step and the range of values for the first derivative of the corresponding physiological data change representation function at each time step.
[0073] The hierarchy determination unit is used to determine the required hierarchy for trend analysis of physiological data of each type and the trend gradient tolerance of physiological data of the current type at different trend analysis levels, based on the range of values to be judged for physiological data of all calculated samples.
[0074] Based on the above process, the trend change patterns of each type of physiological data under normal conditions can be analyzed to reach a certain level, which enriches the trend analysis levels to the greatest extent while ensuring the accuracy of trend analysis.
[0075] Example 4:
[0076] Based on Example 3, the hierarchy determination unit includes:
[0077] The first determining subunit is used to determine the range of differences between all adjacent second derivative values of the physiological data of each calculated sample when the commonality of the range of values to be judged for the same type of physiological data of all calculated samples is not less than the commonality threshold (i.e., the commonality judgment threshold used to determine whether to continue to a new range of values to be judged). This is done until the commonality of the range of values to be judged for the current type of physiological data of all calculated samples is less than the commonality threshold. Then, the value of the number of judgments executed in the previous execution is used as the value of the level required for trend analysis of the current type of physiological data, and the range of values to be judged in each judgment process executed in the previous and previous executions is used as the trend gradient tolerance value of the current type of physiological data at the corresponding trend analysis level.
[0078] The second determining subunit is used to treat the value of the number of judgments executed in the previous execution as the value of the level required for trend analysis of the current type of physiological data when the commonality of the range of values to be judged for the same type of physiological data in all calculated samples is less than the commonality threshold, and to treat the range of values to be judged in the previous execution as the trend gradient tolerance value of the current type of physiological data at the corresponding trend analysis level.
[0079] The commonality of the range of values to be judged for the same type of physiological data in all calculated samples is calculated as follows:
[0080] Calculate the number of times each time step exists in the range of values to be judged for the same type of physiological data in all calculated samples within the range of values to be judged for the current type of physiological data. The ratio of the number of existences to the total number of values to be judged for the current type of physiological data is taken as the existence ratio. The mean of the existence ratios of each value to be judged for all time steps is taken as the total existence ratio of the corresponding value to be judged. The mean of the total existence ratios of all values to be judged for the current type of physiological data is taken as the commonality of the value ranges to be judged for the current type of physiological data in all calculated samples.
[0081] Based on the above iterative process, the commonality of the range of values to be judged under the same trend analysis level of the same physiological data of different calculation samples is introduced as an indicator, which serves as the limit threshold for iterative exploration of trend change patterns. This enables in-depth exploration of the limit of trend change analysis levels for each type of physiological data, and enriches the trend analysis levels for each type of physiological data to the greatest extent.
[0082] Example 5:
[0083] Based on Example 1, the data denoising submodule includes:
[0084] The corresponding derivative unit is used to perform corresponding derivatives on the physiological data change representation function based on the required level of trend analysis for each type of physiological data, to obtain multiple derivative functions of different orders (e.g., first-order derivative, second-order derivative, etc.), and to determine the different derivative values of multiple derivative functions at each time step (when the required level of trend analysis is 1, the physiological data change representation function is differentiated once, that is, only its first-order derivative function is obtained, and the first-order derivative function value at each time step is determined; when the required level of trend analysis is 2, the energy-saving data change representation function is differentiated twice, that is, its first-order derivative function and second-order derivative function are obtained, and the first-order derivative function and second-order derivative function values at each time step are determined, and so on).
[0085] The data denoising unit is used to treat the physiological values of different order derivatives that exceed the trend gradient tolerance of the corresponding trend analysis level in the corresponding single non-outlier recording curve (the physiological values of the corresponding time in the corresponding single non-outlier recording curve whose difference from the corresponding order derivative values of two adjacent times both exceed the trend gradient tolerance of the corresponding trend analysis level) as the denoising data to be judged for the current type of physiological data.
[0086] Using the required level of trend analysis and the corresponding trend variation tolerance value for each type of physiological data as the screening criteria, the trend variation tolerance value of the corresponding numerical difference (which may be the difference of physiological values or the difference of values of different derivatives) of adjacent numerical points exceeds the corresponding trend analysis level is selected as the denoising standard. This is then used to screen numerical points and denoise a single type of non-outlier record curve, greatly improving the denoising accuracy.
[0087] Example 6:
[0088] Based on Example 1, the secondary verification module includes:
[0089] The timing determination submodule is used to determine the timing interval of all the data to be denoised in the corresponding type of physiological data (that is, the interval formed by the time of each value in the data to be denoised in the corresponding type of physiological data).
[0090] The commonality calculation submodule is used to calculate the number of times each time point exists in all types of time-series data to be judged and denoised in all time-series intervals (i.e., the total number of time-series intervals of all types of data to be judged and denoised that include the time point currently calculated). Based on the number of existences and the total number of physiological data types, the inter-class commonality of each time point is determined (i.e., the ratio of the number of existences to the total number of physiological data types is used as the inter-class commonality of each time point, and the inter-class commonality is a numerical value that characterizes the degree of commonality of all types of physiological data to be judged and denoised in time series).
[0091] Based on the above process, the inter-class commonality degree, which represents the degree of commonality of the denoised data to be judged in time series for all types of physiological data, is calculated.
[0092] The inter-class verification submodule is used to perform inter-class verification and denoising on the physiological data to be judged for all types of data based on inter-class commonality, so as to obtain the final denoised physiological data.
[0093] Based on inter-class commonality, further inter-class verification and denoising are achieved for all types of physiological data to be denoised.
[0094] Example 7:
[0095] Based on Example 6, the inter-class verification submodule includes:
[0096] The time determination unit is used to treat the time when the inter-class commonality is less than the inter-class commonality threshold as the retained time (the time in the time interval that does not need to be deleted by denoising), and treat the remaining time in all time intervals except the retained time as the time to be denoised (the time in the time interval that needs to be deleted by denoising).
[0097] The final denoising unit is used to remove the physiological values at the time points that need to be denoised from all types of physiological data to obtain the final denoised physiological data.
[0098] The moment when the inter-class commonality degree in the time interval is not less than the inter-class commonality degree threshold is regarded as the moment that needs to be denoised. That is, the part of the data where multiple physiological data change suddenly at the same time (the sudden change data that is judged as suspected noise data) is deleted. In other words, the part of the physiological data that changes suddenly at the same time due to the patient's movement or other physiological activities such as sneezing is deleted, thereby realizing the deletion of abnormal physiological data in the physiological data.
[0099] Example 8:
[0100] Based on Example 1, the model building module includes:
[0101] The instance acquisition submodule is used to acquire a preset amount (e.g., 10,000) of denoised physiological data of various types and corresponding professional depression assessment values (i.e., depression levels assessed by professionals based on physiological data or other information such as the facial expressions and lifestyle of the person whose physiological data is being collected).
[0102] The sample extraction submodule is used to treat the temporal fluctuation characteristics of all denoised physiological data in each evaluation instance as a single sample.
[0103] The model building submodule is used to divide all samples into training samples and test samples (e.g., in an 8:2 ratio), and to train and test based on the training samples and test samples to obtain a depression level analysis model.
[0104] Among them, temporal fluctuation characteristics include, but are not limited to, extreme values of individual types of physiological data and time-varying functional representations.
[0105] Example 9:
[0106] Based on Example 1, it also includes:
[0107] The warning prompt generation module is used to generate warning prompt information based on the user's current depression level value (that is, preset warning prompt information corresponding to different depression levels. The preset warning prompt information may include prompts to exercise or take a walk, prompts to communicate with others, or prompts to be displayed on other devices to communicate with the tested user).
[0108] The warning information display module is used to display warning information in a single-end (sent to and displayed on a single communication end of the user's communication end or the communication end of the user's relatives and friends) or in a multi-end (sent to and displayed on multiple communication ends of the user's communication end or the communication end of the user's relatives and friends).
[0109] Furthermore, based on Example 1, subsequent maintenance and management of users' mental health were implemented, further improving the intelligence level of the depression level analysis system.
[0110] Example 10:
[0111] Based on Example 9, the warning information display module includes:
[0112] The instruction receiving submodule is used to receive the user's display permission setting instruction (i.e., the instruction input by the user to set the permission range for sharing the current depression level value). The display mode (including single-end display and multi-end display) is parsed from the display permission restriction instruction.
[0113] The first display submodule is used to display the warning information on the corresponding single end when the display mode is single-end display;
[0114] The second display submodule is used to determine the target multi-terminals (i.e., multiple communication terminals that are authorized to receive and know the user's current depression level value) corresponding to the display permissions when the display mode is multi-terminal display, and to display the warning information to the corresponding target multi-terminals;
[0115] The system has enabled the sharing and display permissions for depression level analysis results, implemented privacy permission settings for depression level analysis results, and further improved the intelligence level of the depression level analysis system.
[0116] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A device for analyzing the degree of depression based on machine learning algorithms and physiological data, characterized in that, include: The initial verification module is used to acquire multiple types of physiological data from users in real time, perform self-verification on each type of physiological data, and obtain denoised data to be judged for each type of physiological data. The secondary verification module is used to perform inter-class verification denoising on all types of physiological data to be judged, and obtain the final denoised physiological data. The model building module is used to build a depression severity analysis model based on machine learning algorithms. The severity determination module is used to input the temporal fluctuation characteristics of the final denoised physiological data into the depression severity analysis model to obtain the user's current depression severity value; The initial verification module includes: The outlier denoising submodule is used to generate multiple single-category physiological data recording curves for users based on real-time acquired physiological data of each category, and to remove outliers in the single-category physiological data recording curves to obtain single-category non-outlier recording curves. The curve representation submodule is used to generate a physiological data change representation function that represents the physiological data changes over time in a single non-outlier recording curve. The hierarchy determination submodule is used to determine the hierarchy required for trend analysis of physiological data of each type and the trend variability tolerance of physiological data of the current type at different trend analysis hierarchy. The data denoising submodule is used to determine the denoised data of the current type of physiological data in a single non-outlier recording curve based on the physiological data change representation function of each type of physiological data, the corresponding trend analysis level required, and the trend gradient tolerance at different trend analysis levels. The secondary verification module includes: The temporal determination submodule is used to determine the temporal interval of all data to be denoised in the corresponding category of physiological data; The commonality calculation submodule is used to calculate the number of times each time point exists in all types of denoised data to be judged in the time interval of all types of data. Based on the number of existences and the total number of types of physiological data, the inter-class commonality of each time point is determined. The inter-class verification submodule is used to perform inter-class verification and denoising on the physiological data to be judged for all types of data based on inter-class commonality, so as to obtain the final denoised physiological data. The inter-class validation submodule includes: The time determination unit is used to treat the time when the inter-class commonality is less than the inter-class commonality threshold as the retained time, and treat the remaining time in all time intervals other than the retained time as the time to be denoised. The final denoising unit is used to remove the physiological values at the time points that need to be denoised from all types of physiological data to obtain the final denoised physiological data.
2. The device for analyzing the degree of depression based on machine learning algorithms and physiological data according to claim 1, characterized in that, The hierarchy determines the sub-modules, including: The data representation unit is used to generate a physiological data change representation function for each category of each computational sample based on the physiological data of users in a large number of computational samples that have been de-noised from outliers. The tolerance determination unit is used to determine the range of physiological data differences between adjacent time points in each physiological data of the same category of all calculated samples as the trend gradient tolerance of the first trend analysis level. The range determination unit is used to calculate the first derivative value of the physiological data change representation function at each time step, and determine the range of all first derivative values of the physiological data of each calculation sample as the range of values to be judged. The hierarchy determination unit is used to determine the required hierarchy for trend analysis of physiological data of each type and the trend gradient tolerance of physiological data of the current type at different trend analysis levels, based on the range of values to be judged for physiological data of all calculated samples.
3. The device for analyzing the degree of depression based on machine learning algorithms and physiological data according to claim 2, characterized in that, Hierarchical definition unit, including: The first determining subunit is used to determine the range of differences between all adjacent second derivative values of the physiological data of each calculated sample when the commonality of the range of values to be judged for the same type of physiological data of all calculated samples is not less than the commonality threshold. This range is then used as the new range of values to be judged until the commonality of the range of values to be judged for the current type of physiological data of all calculated samples is less than the commonality threshold. Then, the value of the number of judgments executed in the previous execution is used as the value of the level required for trend analysis of the current type of physiological data, and the range of values to be judged in each judgment process executed in the previous and previous executions is used as the trend gradient tolerance value of the current type of physiological data at the corresponding trend analysis level. The second determining subunit is used to treat the value of the number of judgments executed in the previous execution as the value of the level required for trend analysis of the current type of physiological data when the commonality of the range of values to be judged for the same type of physiological data in all calculated samples is less than the commonality threshold, and to treat the range of values to be judged in the previous execution as the trend gradient tolerance value of the current type of physiological data at the corresponding trend analysis level.
4. The device for analyzing the degree of depression based on machine learning algorithms and physiological data according to claim 1, characterized in that, The data denoising submodule includes: The corresponding derivative unit is used to perform corresponding derivatives on the corresponding physiological data change representation function based on the required level of trend analysis for each type of physiological data, to obtain multiple derivative functions of different orders, and to determine the different derivative values of multiple derivative functions of different orders at each time step. The data denoising unit is used to take the physiological values of different order derivatives of the trend gradient tolerance value that exceed the corresponding trend analysis level in the corresponding single non-outlier record curve as the denoised data to be judged for the current type of physiological data.
5. The device for analyzing the degree of depression based on machine learning algorithms and physiological data according to claim 1, characterized in that, The model building module includes: The instance acquisition submodule is used to acquire a preset amount of denoised physiological data of multiple types and corresponding professional assessment values of depression level for assessment instances. The sample extraction submodule is used to treat the temporal fluctuation characteristics of all denoised physiological data in each evaluation instance as a single sample. The model building submodule is used to divide all samples into training samples and test samples, and to train and test based on the training samples and test samples to obtain a depression level analysis model. Among them, the temporal fluctuation characteristics include the extreme values of physiological data of a single type and the function representation that changes over time.
6. The device for analyzing the degree of depression based on machine learning algorithms and physiological data according to claim 1, characterized in that, Also includes: The early warning generation module is used to generate early warning information based on the user's current depression level. The early warning information display module is used to display early warning information on one end or multiple ends.
7. The device for analyzing the degree of depression based on machine learning algorithms and physiological data according to claim 6, characterized in that, The early warning information display module includes: The instruction receiving submodule is used to receive the user's display permission setting instructions and parse the display mode from the display permission restriction instructions. The first display submodule is used to display the warning information on the corresponding single end when the display mode is single-end display; The second display submodule is used to determine the target multi-devices corresponding to the display permissions when the display mode is multi-device display, and to display the warning information to the corresponding target multi-devices.