Multi-task Classification Model Based On Multi-modal Glioma Data

Abstract

Glioma is a common disease. According to relevant medical research, there is a specific relationship between the appearance of glioma and the genotype of isocitrate dehydrogenase-1 (IDHI). It is also affected by 1p/19q chromosome deletion status. This study uses deep learning techniques to explore the relationship among glioma morphology, IDH1 genotypes and 1p/19q chromosomes based on multi-modal glioma data. We train CNN to obtain the intensity, location and shape of glioma according to MRI images. Taking the features of glioma as input, we use XGBoost to classify the IDH1 genotype and and SVM to classify 1p/19q chromosome status. We find that processing the brain MRI images through CNN can accurately obtain some medical feature information of the glioma, and the accuracy rate of the model is above 0.8. When classifying IDH1 genotype and 1p/19q chromosome status based on these features, we find that the image features of gliomas are more closely related to the IDH1 genotype than to the 1p/19q chromosome status.