State judgment and classification of dynamic functional connectivity patterns
To address the poor performance of traditional clustering algorithms due to the existence of different status connection patterns in dynamic brain network analysis, dynamic functional connection analysis is conducted on patients with neurological diseases in order to find stable connection patterns of the disease to improve classification accuracy.This paper outlines the theoretical background, including brain network construction and analysis.By analyzing the weaknesses of current research, it focuses on the dynamic functional connection calculation and effective feature extraction process.This paper proposes a microstate identification method under cross-validation, and combines it with the dimension compression method based on linear expression to perform stable clustering and feature dimensionality reduction on the original dynamic functional connections; the intra-class distance criterion is introduced to perform intra-group feature optimization under cross-validation.Feature selection, the classification model is optimized and the classification accuracy is improved.A comparison of the current clustering identification state methods in public data sets and mild cognitive impairment identification and classification shows that the proposed method performs better in stable clustering and classification prediction, and in the classification after feature selection.The accuracy reaches 86％, the highest level.
国家科技期刊平台
NSTL
万方数据
