Enhancing Brain Tumor Diagnosis: A Comparative Review of Systems with and without eXplainable AI

Abstract

Deep Learning (DL) and computer vision may be used to distinguish between various anatomical features in the human body. As technology has developed, several DL methods have been used to diagnose brain tumors and have shown promise in terms of early diagnosis. Nevertheless, due to their lack of explainability and state-of-the-art (SOTA) accuracy, a smooth integration of these technologies into clinical workflows raises issues. Existing literature reveals that certain studies have obtained SOTA performance by experimenting with cutting-edge technologies and where the model exhibits a black-box nature, several eXplainable Artificial Intelligence (XAI) techniques have been used to resolve their explainability issues. This review paper will include an overview of brain tumors, a review of the recent research publications that involve XAI, and which do not, along with the techniques used and their performances. A comparative analysis will be presented to gather ideas on the strengths and weaknesses while narrowing down the limitations to discover research gaps which need to be addressed. Finally, key findings from the review which will be followed by concluding remarks and a possible future scope will be included at the end.