CN-122004858-A - Mood disorder assessment system with multi-level feature fusion

CN122004858ACN 122004858 ACN122004858 ACN 122004858ACN-122004858-A

Abstract

The invention provides a multi-level feature fusion mood disorder assessment system which comprises a data acquisition unit, an electroencephalogram feature extraction unit, a multi-level feature extraction unit and a feature fusion and classification unit, wherein the data acquisition unit is used for acquiring electroencephalogram signals of a plurality of brain regions of a patient to be assessed, the electroencephalogram feature extraction unit is used for extracting electroencephalogram features corresponding to the electroencephalogram signals of each brain region of the patient to be assessed, the multi-level feature extraction unit is constructed and obtained based on symptom features and corresponding electroencephalogram features of a plurality of testees and used for extracting electroencephalogram feature latent variables of a plurality of levels of brain regions of the patient to be assessed to obtain a multi-level brain feature latent variable, and the feature fusion and classification unit is used for carrying out classification prediction based on the electroencephalogram feature latent variables to obtain mood disorder assessment results of the patient to be assessed. The invention solves the problem that the mood disorder assessment system in the prior art adopts a single feature extraction and learning strategy, has no constraint of symptom information, and causes the limited identification capability of the model on mood disorders of different abnormal function types.

Inventors

WANG GANG
ZHANG LING
WANG YINGTAN
ZHAO XIXI
CUI YI
WANG BIN
ZHAO TONG
YAN YUXIANG
SHA SHA
LI KE
REN YANPING
LI HONGMIN

Assignees

灵心慧智医学科技(北京)有限公司

Dates

Publication Date: 20260512
Application Date: 20251225

Claims (10)

1. A mood disorder assessment system with multi-level feature fusion, comprising: The data acquisition unit is used for acquiring brain electrical signals of a plurality of brain areas of the patient to be evaluated; The electroencephalogram feature extraction unit is used for extracting electroencephalogram features corresponding to electroencephalogram signals of each brain region; The multi-level feature extraction unit is constructed based on symptom features and corresponding brain electrical features of a plurality of testees and is used for extracting brain electrical feature latent variables of a plurality of levels of brain electrical feature latent variables of brain regions of a patient to be evaluated to obtain the brain electrical feature latent variables of the plurality of levels; and the characteristic fusion and classification unit is used for carrying out classification prediction based on the electroencephalogram characteristic latent variable to obtain a mood disorder evaluation result of the patient to be evaluated.
2. The multi-level feature fusion mood disorder assessment system as recited in claim 1, wherein the electroencephalogram corresponding to the electroencephalogram includes a band energy class feature, a connectivity feature, an amplitude phase coupling feature, a time-frequency spectrum feature, a micro-state feature, a central rhythm feature, and a VAEEG feature.
3. The mood disorder assessment system as recited in claim 2, wherein the extracting brain electrical characteristics corresponding to brain electrical signals of each brain region comprises: Calculating the signal power of each frequency band through band-pass filtering or power spectrum density estimation to obtain the frequency band energy characteristics of each frequency band; calculating the synchronicity of different brain signals in frequency or phase by a coherence and phase locking method to obtain the connectivity characteristic; respectively extracting low-frequency phase and high-frequency amplitude signals by using Hilbert transformation, and calculating the coupling strength between the low-frequency phase and the high-frequency amplitude signals by using a modulation index to obtain the amplitude-phase coupling characteristic; obtaining a dynamic spectrum of signal power changing along with time and frequency by a time-frequency analysis method, and obtaining the time-frequency spectrum characteristic based on the dynamic spectrum; Clustering the multichannel electroencephalogram at GFP peak time to obtain a plurality of template micro states, and quantifying the duration, coverage rate and conversion rule of each state by reverse fitting to obtain the micro state characteristics; obtaining the central rhythm feature by calculating a change in power of a specific frequency band of the sensory-motor cortex during a motor task relative to a baseline period; extracting potential characterization of each 1s brain electricity in the 1-30Hz frequency band by using VAEEG pre-training models to obtain the VAEEG characteristics.
4. The mood disorder assessment system as recited in claim 1, wherein the multi-level feature extraction unit comprises: The primary symptom guiding latent variable extraction module is used for extracting a plurality of primary electroencephalogram characteristic latent variables which are most relevant to global symptom characteristics and correspond to each brain region and whole brain ; The secondary symptom-oriented specific latent variable extraction module is used for extracting secondary electroencephalogram characteristic latent variables containing symptom specificity corresponding to each brain region and whole brain ; The three-level subtype fusion latent variable extraction module is used for extracting the secondary electroencephalogram characteristic latent variable Clustering to obtain multiple brain electrophysiology subtypes corresponding to each brain region Feature extraction and fusion are carried out based on each electroencephalogram physiological subtype to obtain three-level electroencephalogram feature latent variables fusing symptom subtype features 。
5. The system for assessing the mood disorder with multi-level feature fusion according to claim 4, wherein the primary symptom-oriented latent variable extraction module comprises a plurality of primary global regression models corresponding to the number of brain regions, model parameters of the primary global regression models are obtained through partial least square iterative training based on global symptom features corresponding to a plurality of testees and brain electrical features corresponding to each brain region, and the model parameters comprise an brain electrical feature load matrix Model weight matrix 。
6. The multi-level feature fusion mood disorder assessment system as recited in claim 5, wherein the primary global regression model is extracted to obtain primary symptom-oriented latent variables for each brain region by: s1, normalizing each brain electrical characteristic and initializing Make the following And the other parts of the main body are respectively connected with the main body, Representing an electrical brain characteristic X of the H updated running with the latent variable dimension H as a target; S2, will And Multiplying to obtain latent variable Expressed as: M is the number of samples; S3, updating the input characteristics of the next round of brain electricity based on the load matrix, wherein the input characteristics are expressed as follows: ; s4, slave To the point of And S2-S3 is sequentially executed to obtain first-level symptom guiding latent variables corresponding to each brain region and the whole brain.
7. The multi-level feature fusion mood disorder assessment system as recited in claim 5, wherein the secondary symptom-oriented specific latent amount extraction module comprises a plurality of secondary symptom-specific fit models corresponding to a number of symptom features The second order symptom-specific fitting model Based on the first-level electroencephalogram characteristic latent variable And the corresponding symptom characteristics are obtained through iterative training of a partial least square method and are used for reasoning the secondary symptom specific latent variables corresponding to the symptom characteristics 。
8. The multi-level feature fusion mood disorder assessment system as recited in claim 7, wherein the secondary symptom-specific fit model The method is constructed by the following steps: Brain characteristic load matrix based on corresponding brain regions and whole brain First-level brain electrical characteristic latent variable for each brain region and whole brain Performing inverse transformation to obtain reconstructed original features ; Obtaining first-level residual error characteristics corresponding to each brain region and the whole brain based on the original brain electrical characteristics corresponding to each brain region and the whole brain and the difference value of the reconstructed original characteristics ; For each symptom feature, the first-order residual feature is utilized respectively Obtaining a corresponding secondary symptom specificity fitting model through partial least square training 。
9. The multi-level feature fusion mood disorder assessment system as recited in claim 5, wherein the three-level subtype fusion latent variable extraction module comprises: the tertiary subtype clustering module is used for being based on the secondary electroencephalogram characteristic latent variable Clustering to obtain multiple brain electrophysiology subtypes corresponding to each brain region and whole brain And posterior probability of each subtype ; The latent variable extraction module under the view angle of the sub-symptom subtype is used for reasoning and obtaining subtype latent variables based on the PLS projection model of each pre-trained electroencephalogram physiological subtype ; Subtype fusion module under the view angle of sub-symptoms for fusing posterior probability of each electroencephalogram physiological subtype Converting into deterministic weight, and obtaining the three-level brain electrical characteristic latent variable based on the deterministic weight 。
10. The mood disorder assessment system as recited in claim 4, wherein the symptom features are derived based on a mood disorder-related assessment scale, and wherein a mean of symptom features corresponding to each subject is used as the global symptom feature.

Description

Mood disorder assessment system with multi-level feature fusion Technical Field The invention belongs to the technical field of medical informatics, and particularly relates to a mood disorder assessment system with multi-level feature fusion. Background The electroencephalogram is a common measurement method for recording the electrical activity generated by the cortex, has the characteristics of high resolution, noninvasive property and low cost, and can reflect the neural activity of various functional areas of the human brain, so that the electroencephalogram has application in brain disease diagnosis, motor recovery and nervous system disease evaluation. In recent years, in the fields of psychology and cognitive neuroscience, research on the basis of electroencephalogram exploration into mood disorders is also on the rise However, the current mood disorder assessment system based on electroencephalogram is generally based on a single feature extraction and learning strategy, and cannot fully capture the unique functional modes of each brain region, so that the model has limited capacity of identifying mood disorders of different abnormal function types. In addition, when the existing mood disorder assessment system is used for extracting the characteristics of the brain electrical characteristics, the lack of constraint of clinical symptom dimension can lead the model to study the characteristics with differentiation degree, and information hidden in the electrophysiological index and related to diseases is ignored, so that judgment of the model on other diseases with similar electrophysiological index distribution but different symptoms is affected. Disclosure of Invention In view of the above analysis, the present invention aims to provide a mood disorder assessment system with multi-level feature fusion, which is used for solving the problem that in the prior art, a single feature extraction and learning strategy is adopted, and no constraint of symptom information exists, so that the mood disorder identification capability of a model for different abnormal function types is limited. The aim of the invention is mainly realized by the following technical scheme: In one aspect, the present invention provides a mood disorder assessment system with multi-level feature fusion, comprising: The data acquisition unit is used for acquiring brain electrical signals of a plurality of brain areas of the patient to be evaluated; The electroencephalogram feature extraction unit is used for extracting electroencephalogram features corresponding to electroencephalogram signals of each brain region; The multi-level feature extraction unit is constructed based on symptom features and corresponding brain electrical features of a plurality of testees and is used for extracting brain electrical feature latent variables of a plurality of levels of brain electrical feature latent variables of brain regions of a patient to be evaluated to obtain the brain electrical feature latent variables of the plurality of levels; and the characteristic fusion and classification unit is used for carrying out classification prediction based on the electroencephalogram characteristic latent variable to obtain a mood disorder evaluation result of the patient to be evaluated. Further, the electroencephalogram characteristics corresponding to the electroencephalogram signals comprise frequency band energy characteristics, connectivity characteristics, amplitude phase coupling characteristics, time spectrum characteristics, micro state characteristics, central rhythm characteristics and VAEEG characteristics. Further, the extracting of the electroencephalogram features includes: Calculating the signal power of each frequency band through band-pass filtering or power spectrum density estimation to obtain the frequency band energy characteristics of each frequency band; calculating the synchronicity of different brain signals in frequency or phase by a coherence and phase locking method to obtain the connectivity characteristic; respectively extracting low-frequency phase and high-frequency amplitude signals by using Hilbert transformation, and calculating the coupling strength between the low-frequency phase and the high-frequency amplitude signals by using a modulation index to obtain the amplitude-phase coupling characteristic; obtaining a dynamic spectrum of signal power changing along with time and frequency by a time-frequency analysis method, and obtaining the time-frequency spectrum characteristic based on the dynamic spectrum; Clustering the multichannel electroencephalogram at GFP peak time to obtain a plurality of template micro states, and quantifying the duration, coverage rate and conversion rule of each state by reverse fitting to obtain the micro state characteristics; obtaining the central rhythm feature by calculating a change in power of a specific frequency band of the sensory-motor cortex during a motor task relative to a