Clinical Adoption of CAD: Exploration of the Barriers to Translation through an Example Application

Abstract

Computer aided diagnosis (CAD) software is not yet widely used in clinic. This paper aims to identify possible reasons why. Firstly, the technical maturity of CAD is explored through analysis of diagnostic accuracy metrics in one example application, the automated classification of Ioflupane I123 (DaTSCAN) images. Software is developed for image classification based on well- established eigenimage techniques. Using a publicly available database of images an area under the Receiver Operator Curve (AUROC) of 0.980 is achieved.Given these impressive results the main blockage to clinical adoption, both in DaTSCAN classification and potentially in other applications, is likely to relate to wider issues. These are explored with reference to the demands of the National Institute for Health and Care Excellence (NICE) evaluation processes. It is postulated that in order to enable wider adoption a greater focus on proving the safety, efficacy and cost effectiveness of CAD may be required.