Authors: Gargi Mishra, Shankar Thalla
Abstract: The implementation of Artificial Intelligence (AI) algorithms in clinical decision support systems (CDSS) and biomedical diagnostics has already shown its impressive potential for positive patient impact. At the same time, there are several challenges that need to be addressed when it comes to the deployment of advanced machine learning models because of their 'black box' characteristics that cannot satisfy the needs of clinical practice where transparency is vital for trust and validation of generated advice. This paper introduces the concept of Explainable Artificial Intelligence (XAI) for application in clinical diagnostics through the use of both post-hoc explanation methods and inherently interpretable models. We offer an approach that includes using fuzzy probabilistic trees in order to create an interpretable model along with mapping of biomarkers activation to explain a generated solution visually. Evaluation performed on multimodal biomedical data shows that explanation-enabled models can provide comparable diagnostics accuracy while providing actionable explanations that help to increase clinician acceptance by 23.7%.
DOI: http://doi.org/10.5281/zenodo.22724968
