Survival Classification of High-Grade Gliomas using an Interpretable 3D Multi-modal System

The automatic survival classification of High-Grade Gliomas (HGGs) is a laborious task due to the complexity of the gliomas’ morphology, size and location. Most of the available deep learning-based techniques exploit single path architectures which provide limited high-dimensional feature extraction ability and classification accuracies. Moreover, the lack of interpretability analysis of the deep learning systems limits its real time deployment in the CAD-based architectures. Therefore, to overcome these architectural challenges, this research works presents a 3D multi-modal system with interpretation evaluation to categorize HGG subjects into three pre-defined survival classes. The model is equipped with four parallel branches for each modality to extract high-dimensional survival features and in the fifth pathway the modalities are also concatenated as channels. The experimental evaluation is performed on Multimodal brain tumor segmentation dataset (BraTS 2020) dataset and the proposed 3D architecture has achieved high values of accuracy, sensitivity, and specificity in comparison to its counter-parts. Keywords: Survival, Classification, Deep Learning, High-Grade Gliomas, Interpretation.