DGIResNet-based multi-gene classification method for adult diffuse glioma using MRI

Predicting the histological grading and genetic biomarkers of gliomas using medical imaging and deep learning is a highly challenging task with significant importance for providing personalized treatment and survival prediction. Previous studies have primarily focused on predicting the status of a single or two genes. This paper firstly proposes a new end-to-end network framework, GIResNet, to simultaneously predict the status of five genotypes. Our model, based on the ResNet architecture, optimizes channel and spatial scales by concatenating T1C and T2F images. Compared to the original ResNet, our approach demonstrates higher lesion attention and improvements in various metrics, showcasing strong potential for practical applications.