On Identifying Effective Investigations with Feature Finding Using Explainable AI: An Ophthalmology Case Study

Rathika Suresh Kumar, Kelvin Zhenghao Li, Si Yin Charlene Chia, Li Rong Wang, Vivien Cherng Hui Yip, Wei Kiong Ngo, Yih Yng Ng, Xiuyi Fan

Abstract: Over-investigation is a longstanding challenge in the contemporary healthcare system. Employing feature selection techniques from Explainable AI (XAI), we conceptualise medical investigation decisions as a “feature finding” problem in machine learning, utilising XAI feature attribution to select the most effective investigations for each patient. Focused on ophthalmology, our research applies this framework to effectively identify investigations for diagnosing eye conditions on an individual basis. Our results demonstrate the algorithm’s proficiency in accurately identifying recommended investigations that align with clinical judgment. Our contributions include modelling the selection of optimal medical investigations as a feature finding problem and introducing an algorithm for computing optimal investigations.