Perspectives on Assurance Case Development for Retinal Disease Diagnosis Using Deep Learning

We report our experience with developing an assurance case for a deep learning system used for retinal disease diagnosis and referral. We investigate how an assurance case could clarify the scope and structure of the primary argument and identify sources of uncertainty. We also explore the need for an assurance argument pattern that could provide developers with a reusable template for communicating and structuring the different claims and evidence and clarifying the clinical context rather than merely focusing on meeting or exceeding performance measures.