A Review of Interpretable Deep Learning for Neurological Disease Classification
Rahul Katarya, Parth Sharma, Nishant Soni, Prithish Rath · 2022 8th International Conference on Advanced Computing and Communication Systems (ICACCS) · 2022
While deep learning models have been able to boast high accuracies for healthcare applications, their actual clinical use has been limited due to their black-box nature. To tackle this problem, interpretability or explainability methods are used to help provide reasoning behind the decisions of a deep learning model as well as to gain insight into their reasoning process. This is necessary due to the high-stakes nature of healthcare applications, in which every decision made must be well informed. In this paper, we showcase the various techniques that are used for interpretability in neurological disease classification. We discuss the advantages and drawbacks of various methods used to generate explanations and also discuss how to evaluate these explanations. Finally, we also discuss some of the flaws in evaluation techniques and the future directions for this field.