MDD early dynamic diagnosis method based on multi-modal attention network
By dynamically setting the MDD early diagnosis cycle through a multimodal attention network model, the subjectivity and timeliness problems of MDD early diagnosis in existing technologies are solved, the setting of individualized diagnosis cycles is realized, and the accuracy and clinical practicality of MDD diagnosis are improved.
Patent Information
- Application Number
- CN202510777091.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-11
- Publication Date
- 2025-09-19
AI Technical Summary
Existing technologies are highly subjective and have poor timeliness in the early diagnosis of MDD. The fixed diagnostic interval model can easily lead to misdiagnosis or over-treatment, and no mechanism for setting an individualized diagnostic cycle has been established.
A method based on a multimodal attention network is adopted to integrate multimodal data such as blood, neuroimaging and physiological signals, use the multimodal attention network diagnostic model, dynamically set the individualized diagnostic cycle, and combine the hierarchical attention mechanism and neural network model to perform early diagnosis of MDD.
It significantly improves the accuracy and timeliness of early diagnosis of MDD, enhances the reliability of clinical decision-making and efficiency of resource allocation, and provides accurate individualized prevention and control plans.
Smart Images
Figure CN120674037A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of MDD diagnosis, and more particularly, to an early dynamic diagnosis method for MDD based on a multimodal attention network. Background Art
[0002] In the field of mental illness diagnosis and treatment, Major Depressive Disorder (MDD), one of the mental disorders with the highest disability rate worldwide, still faces significant challenges in its early and accurate diagnosis. Traditional diagnostic methods rely heavily on scale assessments and clinical interviews, which have core flaws such as strong subjectivity and poor timeliness. Although the introduction of neuroimaging technologies (such as fMRI and sMRI) and biomarker detection (such as inflammatory factors and neurotrophic factors) has significantly improved objectivity, the existing technical system still has the following key limitations:
[0003] During the early diagnosis of MDD, some patients may not present with prominent symptoms, leading to similarities between their presentations and those of other early psychiatric disorders. This leads to significant overlap in diagnostic data, necessitating long-term diagnosis and often employing fixed diagnostic intervals (e.g., every six months), without a mechanism for individualized diagnostic cycles. This one-size-fits-all approach can lead to two risks: delayed intervention for rapidly progressing patients and excessive medical burdens for those in stable stages.
[0004] In order to break through the above technical bottlenecks, the present invention proposes an early dynamic diagnosis method for MDD based on a multimodal attention network. Summary of the Invention
[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide an early dynamic diagnosis method for MDD based on a multimodal attention network.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] The early dynamic diagnosis method of MDD based on multimodal attention network is as follows:
[0008] Whenever a new target diagnosis user appears, a set of modal data of the target diagnosis user is generated to determine the MDD early diagnosis salience value of the target diagnosis user;
[0009] Determine the diagnosis cycle for the target diagnosis user based on the MDD early diagnosis salience value;
[0010] During each diagnostic cycle, the MDD early diagnosis salience value of the target diagnosis user is determined, and the MDD diagnosis decision value of the target diagnosis user is generated synchronously. Based on the comparison result between the MDD diagnosis decision value and the threshold, it is diagnosed whether the target diagnosis user has early MDD.
[0011] Furthermore, the process of determining the MDD early diagnosis salience value of the target diagnosis user is as follows: integrating the various modal data sets into a full set of modal data, importing the full set of modal data into a multimodal attention network diagnostic model for early diagnosis of various types of mental illnesses, and deriving the early diagnosis value of various types of mental illnesses.
[0012] Furthermore, the diagnosis period for the target diagnosis user is determined based on the MDD early diagnosis salience value: the MDD early diagnosis salience value of the target diagnosis user is marked as SYC, and the range of each MDD early diagnosis salience value is set to correspond to a diagnosis period, and the range of the MDD early diagnosis salience value is [0, SY1], (SY1, SY2], …, (SYC-1, SYC], and the diagnosis period includes diagnosis period 1, diagnosis period 2, …, diagnosis period C-1, and diagnosis period C, wherein the period interval length of diagnosis period 1 is less than the period interval length of diagnosis period 2, …, less than the period interval length of diagnosis period C-1, and less than the period interval length of diagnosis period C.
[0013] Furthermore, the process of determining the MDD early diagnosis salience value of the target diagnosis user is as follows: obtaining the early diagnosis value of each type of mental illness for the target diagnosis user, calculating the MDD diagnosis distinction value, summing up the mean of all MDD diagnosis distinction values, calculating the MDD diagnosis distinction mean, multiplying the MDD early diagnosis value by the MDD diagnosis distinction mean, and calculating the MDD early diagnosis salience value of the target diagnosis user.
[0014] Furthermore, the process of obtaining the MDD diagnostic distinction value is as follows: the early diagnosis value of MDD is selected, and the absolute difference between the early diagnosis value of MDD and the early diagnosis values of other types of mental illnesses is calculated to calculate the MDD diagnostic distinction value.
[0015] Furthermore, the process of generating the MDD diagnostic decision value of the target diagnosis user is as follows: all MDD early diagnosis salient values of the target diagnosis user are obtained, a rectangular coordinate system is constructed with time as the X-axis and the MDD early diagnosis salient value as the Y-axis, the MDD dynamic salient value and the average diagnostic salient slope are further calculated, the ratio of the MDD dynamic salient value and the average diagnostic salient slope is calculated, and the MDD diagnostic decision value of the target diagnosis user is calculated.
[0016] Furthermore, the process of obtaining the MDD dynamic highlight value is as follows: each MDD early diagnosis highlight value of the target diagnosis user is marked in the rectangular coordinate system in the form of coordinate points, multiple coordinate points are marked, and every two adjacent coordinate points are connected to obtain a diagnostic highlight continuous line, and the slope of each diagnostic highlight continuous line is obtained. The absolute sum of the slopes of all diagnostic highlight continuous lines is calculated to calculate the average diagnostic highlight slope.
[0017] Furthermore, the process of obtaining the average diagnostic highlight slope is as follows: all diagnostic highlight continuous lines form an MDD highlight segment, the two ends of the MDD highlight segment are perpendicular to the X-axis, and the total area of the closed figure formed by the two perpendicular lines, the MDD highlight segment and the X-axis is marked as the MDD dynamic highlight value.
[0018] Furthermore, when the MDD diagnostic decision value of the target diagnosis user is higher than the diagnostic decision threshold, the target diagnosis user is diagnosed with early MDD. When the MDD diagnostic decision value of the target diagnosis user is not higher than the diagnostic decision threshold, the diagnosis cycle is adjusted according to the MDD early diagnosis highlighting value of the target diagnosis user.
[0019] Compared with the prior art, the present invention has the following beneficial effects:
[0020] The method of the present invention uses dynamic fusion of multimodal data, a hierarchical attention mechanism, and a multimodal attention network diagnostic model to perform early diagnosis and analysis of various types of mental illnesses. It deeply analyzes the degree of fit between the target diagnosis user's appearance data and various types of mental illnesses, and further dynamically sets the diagnosis cycle for the target diagnosis user to avoid diagnostic errors caused by one-time diagnosis of multimodal data. It customizes corresponding diagnostic cycles for different target diagnosis users, significantly improving the accuracy, timeliness and clinical practicality of early diagnosis of MDD, further enhancing the reliability of clinical decision-making and resource allocation efficiency, and providing an innovative technical solution for the precise prevention and control of mental illness. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 is a method flow chart of the method of the present invention;
[0022] Figure 2 Flowchart for determining salient values for early diagnosis of MDD. DETAILED DESCRIPTION
[0023] like Figure 1-Figure 2 , an early dynamic diagnosis method for MDD based on multimodal attention network, the method is as follows:
[0024] Whenever a new target diagnosis user appears, various modal data sets of the target diagnosis user are generated (modal data sets are not limited to the following items: blood modal set, neuroimaging modal set, physiological signal modal set, the blood modal set includes IL-6, TNF-α, IL-1β, tryptophan, 5-hydroxyindoleacetic acid (5-HIAA), BDNF, NGF, etc., the neuroimaging modal set includes hippocampal volume, amygdala surface area, prefrontal cortex thickness, resting state functional connectivity (default network FC value), task state activation intensity (such as DLPFC activation value in VFT task), and the physiological signal modal set includes basal skin conductance level (SCL), post-stress peak change (SCR amplitude), heart rate LF, and heart rate HF), and the various modal data sets are integrated into a full modal data set. The full modal data set is imported into a multimodal attention network diagnostic model for early diagnosis of various types of mental illnesses, and the early diagnostic values of various types of mental illnesses are derived. The MDD early diagnosis salience value of the target diagnosis user is further determined, and the diagnosis cycle for the target diagnosis user is determined based on the MDD early diagnosis salience value.
[0025] The early diagnosis of each type of mental illness corresponds to a multimodal attention network diagnostic model. Each multimodal attention network diagnostic model is built based on a neural network model. The types of mental illness are not limited to MDD, BD, and PDD. In the specific implementation method, taking the early diagnosis of MDD as an example, the steps for building a multimodal attention network diagnostic model are as follows: collect multiple full sets of modal data, and each full set of modal data is defined as belonging to a target diagnosis user. Build a neural network model, use the full set of modal data as the basic data, train the built neural network model, design a hierarchical attention architecture, first extract each data in the single modal data set through the self-attention mechanism, and then use cross-attention to fuse multimodal information, and introduce a gating network to dynamically adjust the modal weights. In the model training stage, Bayesian optimization is used to adjust the parameters, combined with spectral normalization and label smoothing to prevent overfitting, and the training process is monitored through 5-fold nested cross-validation. LIME is used to generate local explanations to increase Strong interpretability. In this process, an early diagnosis value is assigned to each full set of modal data. The value range of the early diagnosis value is set between 1 and 20. The size of the early diagnosis value has a clear meaning. The larger the value, the more likely the data in the full set of modal data is to show early symptoms of MDD. Then, multiple full sets of modal data are divided into training set, validation set and test set according to a specific ratio. The specific division ratio is determined to be 60%:20%:20%. Stratified sampling is used to ensure that the DBV distribution of each subset is consistent: the cumulative distribution function (CDF) of the DBV of all samples is calculated, the neural network model is repeatedly trained using the training set, and the performance of the training phase is verified with the help of the validation set. The model parameters are adjusted in time according to the verification results. Hyperparameter tuning, overfitting prevention and training monitoring are used. The final model is evaluated using a test set that did not participate in the training to ensure that the results do not rely on data peeking during the training process. Finally, the multimodal attention network diagnostic model is completed.
[0026] The process of determining the MDD early diagnosis salience value of the target diagnosis user: obtain the early diagnosis value of each type of mental illness for the target diagnosis user, select the early diagnosis value of MDD, calculate the absolute difference between the early diagnosis value of MDD and the early diagnosis values of other types of mental illnesses, calculate the MDD diagnostic distinction value, calculate the sum and mean of all MDD diagnostic distinction values, calculate the MDD diagnostic distinction mean (if the MDD diagnostic distinction mean is 0, set the MDD diagnostic distinction mean to 0.05), multiply the MDD early diagnosis value by the MDD diagnostic distinction mean, and calculate the MDD early diagnosis salience value of the target diagnosis user.
[0027] Determine the diagnosis cycle for the target diagnosis user based on the MDD early diagnosis salience value: mark the MDD early diagnosis salience value of the target diagnosis user as SYC, set the range of each MDD early diagnosis salience value to correspond to a diagnosis cycle, the range of the MDD early diagnosis salience value is [0, SY1], (SY1, SY2], ..., (SYC-1, SYC], the diagnosis cycle includes diagnosis cycle 1, diagnosis cycle 2, ..., diagnosis cycle C-1, diagnosis cycle C, wherein the cycle interval length of diagnosis cycle 1 is less than the cycle interval length of diagnosis cycle 2 < ... < the cycle interval length of diagnosis cycle C-1 < the cycle interval length of diagnosis cycle C.
[0028] During each diagnostic cycle, the MDD early diagnosis salience value of the target diagnosis user is determined, and the MDD diagnostic decision value of the target diagnosis user is generated synchronously. When the MDD diagnostic decision value of the target diagnosis user is higher than the diagnostic decision threshold, the target diagnosis user is diagnosed with early MDD. When the MDD diagnostic decision value of the target diagnosis user is not higher than the diagnostic decision threshold, the diagnostic cycle is adjusted according to the MDD early diagnosis salience value of the target diagnosis user.
[0029] The process of generating the MDD diagnostic decision value of the target diagnosis user is as follows: obtain all the MDD early diagnosis salience values of the target diagnosis user, construct a rectangular coordinate system with time as the X-axis and the MDD early diagnosis salience value as the Y-axis, mark each MDD early diagnosis salience value of the target diagnosis user in the rectangular coordinate system as a coordinate point, mark multiple coordinate points, connect every two adjacent coordinate points to obtain a diagnostic salience continuous line, obtain the slope of each diagnostic salience continuous line, calculate the absolute sum of the slopes of all diagnostic salience continuous lines, calculate the average diagnostic salience slope (if the average diagnostic salience slope is 0, then set the value of the average diagnostic salience slope to 0.1), form an MDD salience line segment from all diagnostic salience continuous lines, draw perpendicular lines from both ends of the MDD salience line segment to the X-axis, mark the total area of the closed figure formed by the two perpendicular lines, the MDD salience line segment and the X-axis as the MDD dynamic salience value, calculate the ratio of the MDD dynamic salience value to the average diagnostic salience slope, and calculate the MDD diagnostic decision value of the target diagnosis user.
[0030] The above method uses dynamic fusion of multimodal data, a hierarchical attention mechanism, and a multimodal attention network diagnostic model to conduct early diagnosis and analysis of various types of mental illnesses. It deeply analyzes the degree of fit between the target diagnosis user's appearance data and various types of mental illnesses, and further dynamically sets the diagnosis cycle for the target diagnosis user to avoid diagnostic errors caused by one-time diagnosis of multimodal data. It customizes corresponding diagnostic cycles for different target diagnosis users, significantly improving the accuracy, timeliness and clinical practicality of early diagnosis of MDD, further enhancing the reliability of clinical decision-making and resource allocation efficiency, and providing an innovative technical solution for the precise prevention and control of mental illness.
[0031] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.
[0032] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.
[0033] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0034] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0035] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0036] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0037] If the functions are implemented in the form of software functional 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 application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0038] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.
Claims
1. An early dynamic diagnosis method for MDD based on a multimodal attention network, characterized by: Here’s how: Whenever a new target diagnosis user appears, a set of modal data of the target diagnosis user is generated to determine the MDD early diagnosis salience value of the target diagnosis user; Determine the diagnosis cycle for the target diagnosis user based on the MDD early diagnosis salience value; During each diagnostic cycle, the MDD early diagnosis salience value of the target diagnosis user is determined, and the MDD diagnosis decision value of the target diagnosis user is generated synchronously. Based on the comparison result between the MDD diagnosis decision value and the threshold, it is diagnosed whether the target diagnosis user has early MDD.
2. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 1, characterized in that: The process of determining the MDD early diagnosis salience value for the target diagnosis user is as follows: integrating various modal data sets into a full set of modal data, importing the full set of modal data into a multimodal attention network diagnostic model for early diagnosis of various types of mental illnesses, and deriving the early diagnosis value of various types of mental illnesses.
3. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 1, characterized in that: Determine the diagnosis cycle for the target diagnosis user based on the MDD early diagnosis salience value: mark the MDD early diagnosis salience value of the target diagnosis user as SYC, set the range of each MDD early diagnosis salience value to correspond to a diagnosis cycle, the range of the MDD early diagnosis salience value is [0, SY1], (SY1, SY2], ..., (SYC-1, SYC], the diagnosis cycle includes diagnosis cycle 1, diagnosis cycle 2, ..., diagnosis cycle C-1, diagnosis cycle C, wherein the cycle interval length of diagnosis cycle 1 is less than the cycle interval length of diagnosis cycle 2 < ... < the cycle interval length of diagnosis cycle C-1 < the cycle interval length of diagnosis cycle C.
4. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 1, characterized in that: The process of determining the MDD early diagnosis salience value of the target diagnosis user: obtain the early diagnosis value of each type of mental illness for the target diagnosis user, calculate the MDD diagnosis distinction value, calculate the sum and mean of all MDD diagnosis distinction values, calculate the MDD diagnosis distinction mean, multiply the MDD early diagnosis value by the MDD diagnosis distinction mean, and calculate the MDD early diagnosis salience value of the target diagnosis user.
5. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 4, characterized in that: The process of obtaining the MDD diagnostic discrimination value is as follows: the early diagnosis value of MDD is selected, and the absolute difference between the early diagnosis value of MDD and the early diagnosis values of other types of mental illnesses is calculated to calculate the MDD diagnostic discrimination value.
6. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 1, characterized in that: The process of generating the MDD diagnostic decision value of the target diagnosis user: obtain all the MDD early diagnosis salience values of the target diagnosis user, construct a rectangular coordinate system with time as the X-axis and the MDD early diagnosis salience value as the Y-axis, further calculate the MDD dynamic salience value and the average diagnostic salience slope, calculate the ratio of the MDD dynamic salience value to the average diagnostic salience slope, and calculate the MDD diagnostic decision value of the target diagnosis user.
7. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 6, characterized in that: The process of obtaining the MDD dynamic salience value is as follows: each MDD early diagnosis salience value of the target diagnosis user is marked in the rectangular coordinate system in the form of coordinate points, multiple coordinate points are marked, and every two adjacent coordinate points are connected to obtain a diagnosis salience continuous line, and the slope of each diagnosis salience continuous line is obtained. The absolute sum of the slopes of all diagnosis salience continuous lines is calculated to calculate the average diagnosis salience slope.
8. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 7, characterized in that: The process of obtaining the average diagnostic salient slope is as follows: all diagnostic salient continuous lines form an MDD salient line segment, perpendicular lines are drawn from both ends of the MDD salient line segment to the X-axis, and the total area of the closed figure formed by the two perpendicular lines, the MDD salient line segment, and the X-axis is marked as the MDD dynamic salient value.
9. The method for early dynamic diagnosis of MDD based on a multimodal attention network according to claim 1, characterized in that: When the MDD diagnostic decision value of the target diagnosis user is higher than the diagnostic decision threshold, the target diagnosis user is diagnosed with early MDD. When the MDD diagnostic decision value of the target diagnosis user is not higher than the diagnostic decision threshold, the diagnosis cycle is adjusted according to the MDD early diagnosis highlight value of the target diagnosis user.