CN-121999988-A - Depression recurrence risk intelligent early warning method, system and equipment based on multi-mode data and artificial intelligence

CN121999988ACN 121999988 ACN121999988 ACN 121999988ACN-121999988-A

Abstract

The application discloses an intelligent depression recurrence risk early warning method, system and equipment based on multi-modal data and artificial intelligence, and relates to the field of depression classification early warning; the method comprises the steps of carrying out dimension reduction treatment on individual multi-mode brain image deviation feature vectors, genetic features, environment and clinical features, inputting the dimension reduction treatment to a pre-trained hierarchical integrated classification model and a pre-trained hierarchical integrated risk early warning model to obtain depression subtype classification labels and risk probabilities, adopting multi-mode feature fusion and hierarchical integrated learning strategies in the pre-training process of the hierarchical integrated classification model and the hierarchical integrated risk early warning model, and carrying out model training and verification by using samples from a plurality of data centers. The application improves the accuracy and individuation degree of classification and early warning, and improves the generalization capability and robustness of the model.

Inventors

LIU CHUNHONG
GUO ZHIPENG
LIU YANG

Assignees

北京市中医药研究所
首都医科大学附属北京中医医院

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. An intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence is characterized by comprising the following steps: Acquiring baseline characteristics, multi-modal brain image characteristics, genetic characteristics, environment and clinical characteristics of a target patient; Based on the baseline characteristic and the multi-modal brain image characteristic, performing individual deviation calculation to obtain an individual multi-modal brain image deviation characteristic vector; Performing dimension reduction processing on the individualized multi-modal brain image deviation feature vector, the genetic feature, the environment and the clinical feature, and then respectively inputting the dimension reduction processing to a pre-trained layered integrated classification model and a pre-trained layered integrated risk early warning model to obtain a depression subtype classification label and risk probability; In the pre-training process of the hierarchical integrated classification model and the hierarchical integrated risk early warning model, a multi-mode feature fusion and hierarchical integrated learning strategy are adopted, and samples from a plurality of data centers are used for model training and verification.
2. The intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence according to claim 1, wherein the model adopted for the personalized deviation calculation is one of a Gaussian process regression model, a Bayesian regression model and a quantile regression forest model; when a Gaussian process regression model is adopted, based on the baseline characteristic and the multi-modal brain image characteristic, performing individual deviation calculation to obtain an individual multi-modal brain image deviation characteristic vector, including: selecting one of a plurality of preset brain regions as a current brain region; Invoking a corresponding pre-trained Gaussian process regression model aiming at each modal brain image characteristic by taking the current brain region as a target, and determining a corresponding expected normal value and an expected standard deviation by taking the baseline characteristic as an input; calculating the individuation deviation Z score of the current brain region according to the expected normal value, the expected standard deviation and the corresponding modal brain image characteristics; And combining the individuation deviation Z scores corresponding to all brain areas and all the modal brain image features to determine individuation multi-modal brain image deviation feature vectors.
3. The intelligent pre-training method for risk of depression recurrence based on multimodal data and artificial intelligence according to claim 2, wherein the pre-training process of the gaussian process regression model comprises: Collecting baseline characteristic samples and multi-modal brain image characteristic samples of a plurality of healthy contrasters from a plurality of data centers; Taking the baseline characteristic sample of the healthy control person as input and the multi-mode brain image characteristic sample as output, and training a Gaussian process regression model; And storing the trained Gaussian process regression model and the corresponding healthy crowd distribution parameters for calling.
4. The intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence according to claim 1, wherein the algorithm for performing the dimension reduction processing is one of a stacked self-encoder, a principal component analysis and a linear discriminant analysis.
5. The intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence according to claim 1, wherein the structure of the layered integrated classification model is the same as that of the layered integrated risk early warning model, the training target of the layered integrated classification model is depression subtype classification, and the training target of the layered integrated risk early warning model is risk level early warning; In terms of structure, the hierarchical integrated classification model comprises a plurality of base learners and element learners, wherein the individualized multi-mode brain image deviation feature vector, the genetic feature, the environment and the clinical feature which are subjected to dimension reduction are taken as input features and are respectively input into different base learners, and the output probabilities of all the base learners are taken as new feature vectors and are input into the element learners to obtain a final recognition result.
6. The intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence according to claim 5, wherein the type of the basic learner comprises a Support Vector Machine (SVM) and a random forest, and the type of the meta learner comprises logistic regression and XGBoost.
7. The intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence according to claim 1, wherein the acquisition process of the baseline characteristic, multi-modal brain image characteristic, genetic characteristic, and environmental and clinical characteristics of the target patient comprises: acquiring baseline data, multi-mode brain image data, genetic data and environmental and clinical data of a target patient; sequentially performing standardization processing, registration processing, denoising processing and feature extraction on the multi-modal brain image data to obtain multi-modal brain image features; and sequentially carrying out data standardization and feature extraction on the baseline data, the genetic data and the environmental and clinical data to obtain baseline features, genetic features and environmental and clinical features.
8. The intelligent depression recurrence risk early warning method based on multimodal data and artificial intelligence according to claim 7, wherein the baseline data comprises age, gender, height, weight and education age; the multi-modal brain image data comprises a structural magnetic resonance image, a functional magnetic resonance image and a diffusion tensor image; The genetic data includes gene expression and neurotransmitter distribution; the environment and clinical data comprise scale scoring data and family life information.
9. An intelligent depression recurrence risk early warning system based on multi-modal data and artificial intelligence, to which the depression recurrence risk intelligent early warning method based on multi-modal data and artificial intelligence according to any one of claims 1 to 8 is applied, characterized in that the system comprises: The data acquisition and storage module is used for acquiring baseline characteristics, multi-mode brain image characteristics, genetic characteristics, environment and clinical characteristics of a target patient; the individuation deviation calculation module is used for carrying out individuation deviation calculation based on the baseline characteristic and the multi-modal brain image characteristic so as to obtain an individuation multi-modal brain image deviation characteristic vector; The classification and risk early warning module is used for carrying out dimension reduction treatment on the individualized multi-mode brain image deviation feature vector, the genetic feature, the environment and clinical feature, and then respectively inputting the dimension reduction treatment to a pre-trained hierarchical integrated classification model and a pre-trained hierarchical integrated risk early warning model to obtain a depression subtype classification label and risk probability, wherein the hierarchical integrated classification model and the hierarchical integrated risk early warning model both adopt multi-mode feature fusion and hierarchical integrated learning strategies in the pre-training process, and model training and verification are carried out by using samples from a plurality of data centers; And the interaction module is used for visually displaying and exporting the baseline characteristics, the multi-mode brain image characteristics, the genetic characteristics, the environment and clinical characteristics, the depression subtype classification labels and the risk probability of the target patient.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the multimodal data and artificial intelligence based depression recurrence risk intelligent pre-warning method according to any one of claims 1-8.

Description

Depression recurrence risk intelligent early warning method, system and equipment based on multi-mode data and artificial intelligence Technical Field The application relates to the field of depression classification early warning, in particular to an intelligent early warning method, system and equipment for depression recurrence risk based on multi-mode data and artificial intelligence. Background Major depressive disorder (Major Depressive Disorder, MDD) is a mental disorder with high heterogeneity and a great individual variability in clinical manifestations and therapeutic response. At present, clinical diagnosis and risk assessment mainly depend on subjective scales and lack objective biological markers. Neuroimaging techniques offer the opportunity to explore the neural mechanisms of MDD. However, most of the existing researches adopt a group average comparison method, and huge differences among patients are ignored, so that research results cannot be converted into clinically applicable individual diagnosis tools. Meanwhile, while multi-modal fusion and machine learning methods have been attempted for MDD classification, features of different modalities are often simply stitched, deep, organic fusion cannot be achieved, model performance improvement is limited, and overfitting is easy. Furthermore, most models are built based on single-center data, and the generalization ability drops dramatically in the face of multi-center data for different scanning devices, protocols, and populations. Disclosure of Invention The application aims to provide an intelligent depression recurrence risk early warning method, system and equipment based on multi-mode data and artificial intelligence, which improve the accuracy and individuation degree of classification and early warning and improve the generalization capability and robustness of a model. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the application provides an intelligent depression recurrence risk early warning method based on multi-modal data and artificial intelligence, comprising the following steps: Acquiring baseline characteristics, multi-modal brain image characteristics, genetic characteristics, environment and clinical characteristics of a target patient; Based on the baseline characteristic and the multi-modal brain image characteristic, performing individual deviation calculation to obtain an individual multi-modal brain image deviation characteristic vector; Performing dimension reduction processing on the individualized multi-modal brain image deviation feature vector, the genetic feature, the environment and the clinical feature, and then respectively inputting the dimension reduction processing to a pre-trained layered integrated classification model and a pre-trained layered integrated risk early warning model to obtain a depression subtype classification label and risk probability; In the pre-training process of the hierarchical integrated classification model and the hierarchical integrated risk early warning model, a multi-mode feature fusion and hierarchical integrated learning strategy are adopted, and samples from a plurality of data centers are used for model training and verification. In a second aspect, the present application provides an intelligent early warning system for risk of recurrence of depression based on multi-modal data and artificial intelligence, to which an intelligent early warning method for risk of recurrence of depression based on multi-modal data and artificial intelligence is applied, the system comprising: The data acquisition and storage module is used for acquiring baseline characteristics, multi-mode brain image characteristics, genetic characteristics, environment and clinical characteristics of a target patient; the individuation deviation calculation module is used for carrying out individuation deviation calculation based on the baseline characteristic and the multi-modal brain image characteristic so as to obtain an individuation multi-modal brain image deviation characteristic vector; The classification and risk early warning module is used for carrying out dimension reduction treatment on the individualized multi-mode brain image deviation feature vector, the genetic feature, the environment and clinical feature, and then respectively inputting the dimension reduction treatment to a pre-trained hierarchical integrated classification model and a pre-trained hierarchical integrated risk early warning model to obtain a depression subtype classification label and risk probability, wherein the hierarchical integrated classification model and the hierarchical integrated risk early warning model both adopt multi-mode feature fusion and hierarchical integrated learning strategies in the pre-training process, and model training and verification are carried out by using samples from a plurality of data centers; And the interaction module