Brain State Classification Of Radiologist: A Resting FMRI Study

Vision is the most important way for humans to obtain external information.About 80%of the information is obtained by vision.Human visual recognition ability is of great significance to many aspects such as its growth and learning,and the neural mechanism behind it has been a research hotspot in the academic world.In the field of cognitive neuroscience,visual experts refer to those who can quickly and accurately classify and identify certain types of objects in visual recognition.Their behavior in visual recognition is very stable and better than ordinary people.Therefore,visual experts are often used as the object of studying human visual recognition.However,there is still a big gap in the research on the classification of visual experts and brain representations.How to better define the biological characteristics of visual experts remains to be studied in depth.With the multi-variable method of pattern recognition technology is widely used in neuroimaging data analysis,the advantages of the combination of the both have brought great hope for the brain state classification and representation research of visual experts.For the resting-state functional Magnetic Resonance Imaging data,the pattern recognition method is used to classify the brain state of visual experts and to focus on exploring the the biomarkers and working mechanism of the brain of visual experts.This will have a significant impetus for the research of brain plasticity of visual experts and the research of human growth and learning.The research work and innovation point of this thesis are summarized as follows:First of all,this thesis uses radiologist as visual expert to conduct research.Unlike previous single-scale feature extraction and classification methods,multiscale feature fusion is used for classification research,and explores whether it is possible to define the information representation of the visual expert's brain through an optimized combination of multiscale features.The study collected rs-fMRI data from 62 medical interns,of which radiologist group is 31 interns in imaging department and the normal control group is 31 interns in other departments.Based on the Brainnetome246 atlas,regional average features at three scales of Regional Homogeneity,Amplitude of Low-Frequency Fluctuations,and Degree Centrality are extracted and fused.The wrapped method based on Recursive Feature Elimination is designed to select features.The linear Support Vector Machine classifier and The Leave-One-Out Cross-Validation strategy are used to classify the two groups of samples.The experimental results found that multiscale feature fusion has a better analysis effect on the radiologist's brain state classification than single-scale feature,and can more accurately distinguish the two groups of people.The accuracy,specificity and sensitivity are all 100%.After feature selection,22 most important multiscale features were finally found.Further localization analysis of the brain regions,the results show that radiologists and non-experts are significantly different in the representation of brain regions responsible for visual information processing,semantic processing,long-term memory,and attention.Such as right Middle Temporal Gyrus₂,right Medio Ventral Occipital Cortex₅,right lateral Occipital Cortex₃,right Inferior Temporal Gyrus₁ and right Parahippocampal Gyrus₅.Secondly,based on the above sample data,this thesis proposes a classification method different from the previous local feature analysis.This method uses Functional Connectivity features to conduct classification research,and explores whether the optimal combination of Functional Connectivity features can be used to define the information representation of the brain of visual experts.Based on the Brainnetome246 atlases,the Functional Connectivity features based on the whole brain area were extracted.Least Absolute Shrinkage and Selection Operator method was used to select features and linear Support Vector Machine classifier and Leave-One-Out Cross-Validation strategy were used to classify the two groups of samples.The experimental results found that the Functional Connectivity feature has a significant effect on the analysis of radiologist's brain state classification,and can accurately distinguish the two groups of people.The accuracy,specificity and sensitivity are 91.94%,83.87%and 100%,respectively.After feature selection,8 most important Functional Connectivity features were finally found,involving a total of 14 brain regions.Further analysis shows that radiologists and non-experts are significantly different in the representation of Functional Connectivity in brain regions responsible for visual information processing,long-term memory,and advanced cognition.The Parahippocampal Gyrus region has multiple Functional Connectivity participation.It is a brain region that needs to be focused on.