Clinical decision support system : an explainable AI approach

Abstract

In healthcare, integrating sophisticated artificial intelligence (AI) capabilities with the practical insights required by medical practitioners poses a significant challenge. This study addresses the gap between the capabilities of AI systems and the comprehension of end-users by focusing on the explainability of AI models. The primary objective was to investigate fundamental elements of AI model explainability and data privacy, specifically emphasizing healthcare applications and potential implementations in diverse industries. Utilizing the SHapley Additive exPlanations (SHAP) method, a Clinical Decision Support System (CDSS) was developed, employing Django to build a website that delivers SHAP explanations of models and additional distribution explanation plots. Interactive features were incorporated to enhance user exploration. The system sported efficacy with the Wisconsin Diagnostic Breast Cancer (WDBC) dataset and a model trained on it, along with the Type 1 Diabetes (T1D) Exchange Registry-based model. Case studies were presented to illustrate the tool's usage. Identified limitations and challenges include the need for a better understanding of model architecture, the inclusion of regression models, and addressing version inconsistencies between models and backend. As the tool continues to evolve and integrate future enhancements, it promises to change healthcare decision-making and enhance patient outcomes on a larger scale. This advancement paves the way for a future where transparent and interpretable machine learning models become integral in improving healthcare outcomes for all, without compromising their accuracy.