Developing, purchasing, implementing and monitoring AI tools in radiology: practical considerations. A multi-society statement from the ACR, CAR, ESR, RANZCR & RSNA

Insights into Imaging

Table 4 Purchasing considerations for AI models in radiology

Strategic	Which problem is the AI helping to solve?
	What benefit can be expected from the AI’s usage?
	How much improvement can be expected?
	Are there any risks associated with the AI’s usage? How can those be mitigated?
Regulatory	What is the AI’s intended use?
Regulatory	At which risk category was the AI certified?
Performance	How can the AI’s performance be monitored?
	Can AI failures be detected and reported?
	Is performance on local data comparable to claimed performance?
	Are differences between local data and training data known?
	Does performance vary depending on the imaging device used?
	Does performance vary depending on patient characteristics (gender, ethnicity, etc.)?
Workflow	How is the AI integrated into the radiologist’s workflow?
	Are radiologists biased by the AI’s predictions?
	What training is required for proper usage and bias avoidance?
Technical	How does the AI integrate into local IT infrastructure?
Economic	What is the direct cost of the AI (e.g., licensing)? Which other costs need to be considered (e.g., legal, training, etc.)
Economic	Can return on investment be estimated and monitored?