Brain Network Intelligence Prediction Method Based On Magnetic Resonance Imaging

The brain is the most complex and subtle organ in the human body,responsible for processing all kinds of external information and reflecting it correctly.Tens of billions of interconnected neurons in the brain make up the diversity of human cognitive and behavioural patterns.Intelligence is an important factor in the creation of civilisation and is the driving force behind the growth and progress of the individual,which is why it has been the subject of so much research.Intelligence is all-encompassing,with logical reasoning,thinking,memory and problem-solving skills all falling under the umbrella of intelligence,but its abstract and non-directly measurable nature is a major challenge to research.Non-invasive magnetic resonance imaging(MRI)is an important tool for exploring brain activity and its neural mechanisms,and the complex structure of the brain dictates that multimodal MRI together will provide a better and more comprehensive picture of the brain’s intellectual mechanisms.This paper aims to improve the accuracy of intelligence prediction and to investigate the mechanisms of sex differences in intelligence,structural and functional brain network intelligence.(1)To address the problem of loss of brain network connectivity information caused by not considering the retention of all brain area nodes that send propagation information to the current brain area node in the modeling process of intelligence prediction,this paper proposes an edge-driven graph neural network intelligence prediction method based on shared weights,which preserves the propagation information between all brain areas and the current brain area through a shared weighting strategy.The method fuses brain topological information with graph theory and contains an input layer,an output layer and multiple intermediate layers,where each intermediate layer node value update is completed through three steps: graph convolution calculation,tensor product calculation and feature transformation activation.In order to verify the performance of this method on classification experiments,fluid intelligence regression experiments and gender classification experiments were conducted in this paper.According to the experimental results,the performance of the proposed method on the HCP dataset for both fluid intelligence regression experiments and gender classification experiments was significantly improved compared to other methods.(2)Previous studies on the relationship between individual intelligence and gender have often used linear regression methods,which ignore the topological information of individual brains.In this paper,we investigate the gender differences between individuals’ fluid and crystal intelligence through an edge-driven graphical neural network intelligence prediction method based on shared weights,and conduct two intelligence prediction comparison experiments for males and females to explore the gender differences between males and females in crystal and fluid intelligence.The results of the experiment showed that female intelligence was more easily predicted than male intelligence.During the analysis of the results,an online training module was added to the ws-GPN to conduct an exploration of the mechanisms of gender differences in intelligence.(3)To address the problem that intelligence is currently studied mainly by functional magnetic resonance imaging,there are few prediction studies and comparative studies of diffusion tensor imaging modality,and the results obtained from single modality data prediction cannot fully reveal the brain network mechanism of intelligence,this paper conducts sufficient experiments to investigate the prediction ability of brain structural network and brain functional network in two prediction tasks,namely fluid intelligence and crystal intelligence,and the prediction ability of the two brain networks in each The neurobiological mechanisms of the two brain networks in each intelligence prediction task.The experiments demonstrate that the functional brain network has better predictive power and that the two brain networks are complementary in the intelligence prediction task.Combining the problem that single neuroimaging modality data cannot fully and completely reveal the relationship between brain activity and individual cognitive behaviour,this paper uses a multimodal intelligence prediction model by fusing fmri data and dti data to conduct intelligence prediction experiments.The experimental results show that the multimodal prediction model effectively fuses the two modal brain network data and achieves better prediction performance than unimodal.This paper adopts a graphical neural network-based approach to brain functional connectivity network intelligence prediction,the mechanism of gender differences in intelligence,the investigation of the mechanism of differences in the way brain networks are constructed and multimodal intelligence prediction research,which is important in exploring the brain network mechanisms of both types of intelligence between individuals,understanding and developing the brain comprehensively and accurately,and helping to propose new neural network models inspired by brain structure and function.