The Application Of Clustering-evolutionary Clustering Algorithm In Multimodal Brain Science Data Analysis

Fusing genetic and neuroimaging factors to explore the brain is a frontier issue in the field of brain research,and it is one of the scientific and technological highlights that many countries have invested a lot of manpower and material resources to compete for.Detecting the correlations between genetic data and resting-state functional magnetic resonance imaging data of the brain is an effective way to explain the mysteries of the brain.In this study,the multimodal data of Alzheimer's Disease(AD)and Parkinson's Disease(PD)are fused to construct the correlations between brain regions and genes as the multimodal fusion feature and designing the data analysis model based on the clustering-evolutionary random clustering technology.The main contents are as follows:(1)Clustering-evolutionary random clustering technology is proposed in this study.This technology innovatively combines clustering evolutionary strategy with ensemble learning to improve the performance of ensemble learner in sample recognition and feature selection.Specifically,the method constructs an initial random ensemble model by randomly selecting samples and sample attributes,then uses threshold filtering and hierarchical clustering to carry out multi-level dynamic evolutions of the ensemble learner,and uses the evolved ensemble learner for sample classification and feature extraction.Multilevel clustering evolutions and threshold filtering effectively guarantee the diversity and effectiveness of the base learners in the ensemble model.(2)Clustering-evolutionary Random Forest(CERF)is used to study Alzheimer's disease.In this study,resting-state functional magnetic resonance imaging data and gene data of 40 normal subjects and 38 AD patients are obtained from the ADNI database,and the correlations between genes and brain regions are extracted as sample features.Then,clustering-evolutionary random clustering technology is applied to construct CERF for sample classification and feature extraction.Experiment results show that the CERF model achieves 91.3% accuracy in sample classification,and detects lesion brain regions and risk genes in AD patients.Our study helps to reveal the pathogenesis of AD more comprehensively.(3)Clustering-evolutionary Random Neural Network Ensemble(CERNNE)is used to study Parkinson's disease.In this study,resting-state functional magnetic resonance data and gene data of 55 PD patients and 49 normal subjects are obtained from the ADNI and PPMI databases,and brain region-gene associations are detected to construct sample characteristics.We use multiple neural networks as base learners to construct CERNNE models.In the classification of PD patients and normal people,the effective recognition rate of CERNNE model reaches 88.7%,and the PD-related brain regions and genes are found.Our study may inspire researchers to look for ways to treat and prevent Parkinson's disease from a genetic perspective.