An excellent publication by Daye et al (2022) proposing four key questions:
Who selects the algorithms to implement?
Who will form the AI governing body? (e.g. clinical, research, legal and IT representatives). Will the process be clear and unbiased? Should the governance be radiology-led* vs enterprise-led?
*Radiology has led multiple technological revolutions.
What to consider when evaluating an algorithm for implementation?
Scientific evidence? Inherent Bias? Technical readiness? Workflow impact? Cost? Benefits? Data security risks? Vendor marketing?
How to implement a model in clinical practice?
Where will the integration points lie? How will it access data?
How to monitor and maintain a model after implementation
how will the algorithm be monitored? Which metrics will be used?