Recognising errors in AI implementation in radiology: A narrative review
Nikolaos Stogiannos, Renato Cuocolo, Tugba Akinci D'Antonoli, Daniel Pinto Dos Santos, Hugh Harvey, Merel Huisman, Burak Kocak, Elmar Kotter, Karim Lekadir, Susan Cheng Shelmerdine, Kicky G van Leeuwen, Peter van Ooijen, Michail E Klontzas, Christina Malamateniou
The implementation of AI can suffer from a wide variety of failures. These failures can impact the performance of AI algorithms, impede the adoption of AI solutions in clinical practice, lead to workflow delays, or create unnecessary costs. This narrative review aims to comprehensively discuss different reasons for AI failures in Radiology through the analysis of published evidence across three main components of AI implementation: (i) the AI models throughout their lifecycle, (ii) the technical infrastructure, including the hardware and software needed to develop and deploy AI models and (iii) the human factors involved. Ultimately, based on the identified errors, this report aims to propose solutions to optimise the use and adoption of AI in radiology.