Classification and identification method based on multi-modal multi-site data fusion

Abstract

The invention discloses a classification and identification method based on multi-modal multi-site data fusion. The method comprises the following steps: 1) acquiring multi-site sMRI and R-fMRI data; 2) preprocessing the sMRI data of each site to obtain an sMRI skull stripping brain image and a cerebral cortex surface model; (3) dividing the cerebral cortex into a plurality of different areas, and calculating dissection parameters such as the cortex average thickness, the cortex average surface area and the grey matter volume of each area; 4) inputting the sMRI skull dissection brain image into the ResNet3D deep network model to extract high-dimensional features; 5) fusing the extracted anatomical parameters and high-dimensional features to obtain one-dimensional vectors as sMRI data features; 6) preprocessing the R-fMRI data of each site to obtain a brain grey matter image; the multi-site data has the same or similar data distribution by adopting the method based on low-rank representation, so that the problem of multi-site data isomerism is solved or partially solved, the generalization of a diagnosis model is effectively improved, and actual requirements are better met.

Description

technical field [0001] The invention relates to the technical fields of structural nuclear magnetic resonance image processing, resting state functional nuclear magnetic resonance processing, brain function network analysis, domain self-adaptation, pattern recognition, etc., specifically a classification and recognition method based on multi-modal multi-site data fusion. Background technique [0002] In recent years, the rapid development of medical imaging has provided very important clinical reference value for the analysis of brain imaging data and the observation of brain activity status, which has brought the research of human brain into a new stage. However, due to the lack of brain imaging data, researchers mostly use multi-site data for brain research. [0003] Research on brain diagnosis based on multi-site data can be roughly divided into two categories. The first category is to use only the data of a certain site for model training, and the second category is to d...

AI Technical Summary

Problems solved by technology

However, since different scanner devices have different scanning protocols and scanning parameters, etc., the data distribution of different sites will be different, and there is a problem of data heterogeneity, so the accuracy of the model obtained by using the data training of all sites is low.

At the same time, the data sample size of each site is limited. If only the data of a certain site is used for model training, the obtained model will have the problem of poor generalization ability.

Images