SSDHR classifier network for brain tumor detection using MRI images

Abstract

Brain tumors are the chaotic growth of abnormal brain cells, which is the main cause of cancer worldwide. Symptoms vary depending on tumor size and location, and these tumors can be cancerous or non-cancerous. Early detection and characterization are crucial but challenging due to the brain's complex structure. This study proposes a self-supervised classification deep hierarchical reconstruction network (SSDHR), a combination of generative and discriminative models. This network uses convolutional Kernels of different sizes for processing of input data on separate branches to extract effective discriminative features. A parallel-structured symmetric fusion attention (SFA) module is introduced, combining attention mechanisms for coordinates, time, and frequency to enhance the model's comprehension and feature selection. The SSDHR classifier network achieves the highest accuracy (91.9%) among 10 state-of-the-art classifiers, classifying tumors into 4 classes: pituitary, meningioma, glioma, and no tumor. The model's performance is evaluated using various metrics that demonstrate its effectiveness in brain tumor classification.