ACR DSI values its industry partners and works with them to inform development of clinically relevant AI algorithms around its use cases. In doing so, ACR DSI connects developers with radiology practices that have data sets for algorithm training and testing and provides opportunities for developers to validate their algorithms through the ACR Certify-AI program. ACR DSI maintains directories of the AI models built around its use cases and market availability of these models. ACR DSI also monitors algorithm effectiveness in clinical practice and uses ACR Assess-AI to provide developers with longitudinal algorithm performance data. This data can form the basis for meeting FDA post-market surveillance requirements and for adaptive learning of the algorithm.
DSI operations are led by senior scientists and TOUCH-AI Panels composed of radiologists and experts with backgrounds in data science. Each panel prioritizes use cases according to those that yield the greatest clinical and operational value with attention to the degree to which problems are most amenable to AI solutions. A TOUCH-AI use case includes both an overview and the technical specifications of the clinically-relevant algorithm. TOUCH-AI assists developers by detailing critical content pertinent to an algorithm’s life cycle such as value proposition, common data elements, logical flow charts that inform decision support, and guidelines for annotations.
TOUCH-AI (Technically Oriented Use Cases for Healthcare AI)
Value Proposition: details the algorithm’s value and justification for an expedited review process to regulatory agencies
Common Data Elements: details the input and output variables of the algorithm in the clinical environment and the variables necessary for effective algorithm training
Care Pathways & Flow Charts: offer clinical guidance based on ACR, or other radiological society, criteria for condition classification, characterization, or management. These help developers understand the specific questions an algorithm should reliably address.
Annotation Guidance: guides annotating radiologists to create precise datasets for algorithm training