基于解剖MRI的大脑皮质表面影像组学重建方法研究OA北大核心CSTPCD

Objective:To design a computational method of cortical surface radiomics,to provide rich and reliable local features of brain regions for brain imaging research. Materials and Methods:Based on the T1WI magnetic resonance data sets of 21 groups of repeated measurements of healthy subjects and 222 attention deficit hyperactivity disorder (ADHD)-related subjects,four surface morphological indices including cortical thickness,gray matter volume,mean curvature and cortical surface area were extracted. Using the Desikan-Killiany (DK) brain atlas and spherical local projection,the brain area is flattened from the three-dimensional cortical surface to two-dimensional. Pyradiomics was used to extract 968 two-dimensional radiomics features for each of the four morphological indices. Combining repeated measurement data set and intra-class correlation coefficients (ICC),the ICC value was used as the standard for evaluating radiomics features to comprehensively evaluate the differences in test-retest reliability among different morphological indices,different radiomics feature types and different brain regions. And based on the ADHD dataset,we predict the patient's attention deficit index and hyperactivity index. Results:For different morphological indicators,the radiomics features of gray matter volume and cortical surface area have better reproducibility,and are significantly different from the cortical thickness and average curvature groups (P<0.05). For different types of radiomics features,the first-order features and gray-level co-occurrence matrix features based on cortical thickness showed significant differences from other types of features (P<0.05). For different brain regions,the features extracted from the left and right entorhinal cortex,the left and right temporal poles,and the right frontal pole have lower retest retestability than other regions (P<0.05). However,in general,the brain radiomics features extracted by the surface reconstruction method proposed in this study have high reproducibility (mean ICC>0.76). In the prediction tasks of the two symptom indicators of attention deficit hyperactivity disorder (ADHD),it was found that the left hippocampal gyrus,superior frontal gyrus and superior temporal gyrus were significantly correlated with ADHD symptoms (|r|=0.33-0.52,P<0.05). Conclusions:It is feasible to construct brain radiomics features based on DK brain atlas and surface morphology index. The extracted new features have good repeatability and have certain clinical value in attention prediction and other studies.