New Liability Framework for Artificial Intelligence in Medical Imaging
Legal challenges of next-generation artificial intelligence systems used in medical imaging are addressed, and a five-layer liability model is proposed.
General-purpose and autonomous artificial intelligence systems used in medical imaging highlight discrepancies between existing legal regulations and human-AI workflows. A new five-layer accountability framework is introduced to clarify responsibilities among developers, institutions, and clinicians.
Evolution of Artificial Intelligence in Medical Imaging
Medical imaging was among the first fields to adopt artificial intelligence thanks to limited tasks such as nodule detection or tumor segmentation. However, general-purpose systems capable of processing heterogeneous images and coordinating multi-task operations are transforming this framework.
Five-Layer Liability Framework
An accountability framework consisting of five layers—data, model, orchestration, clinical use, and institutional lifecycle—is proposed. This structure aims to determine where control lies when a system is authorized or audited.
Data and Model Liability
Data liability questions where each input in the workflow comes from and whether it complies with predefined rules. Model liability covers which tasks each model is permitted to perform and its performance conditions.
Orchestration and Clinical Oversight
Large language models or vision models acting as orchestrators choose which tool to call and maintain execution audit logs. In clinical use, artificial intelligence outputs must remain under the meaningful supervision of a competent clinician.
Institutional and Lifecycle Audit
Procurement, integration, staff training, performance monitoring, and update decisions are the responsibility of institutions. Institutions must regularly review audit logs and reauthorize systems following updates.
Insurance and Future-Oriented Approaches
Traditional medical malpractice and product liability policies create uncertainties in artificial intelligence incidents. Insurance models are expected to evolve in line with authorities delegated to artificial intelligence and human intervention capacity.