Artificial intelligence in medicine: a clinician-oriented guide and operational framework to evaluate new AI devices.
Authors
Affiliations (6)
Affiliations (6)
- Laboratoire des Sciences de l'Ingénieur, de l'Informatique et de l'Imagerie (ICUBE), University of Strasbourg, Strasbourg, France.
- Department of Computer Science, University College London, London, UK.
- Department of General Surgery, Hadassah Hebrew University Medical Center, Jerusalem, Israel.
- Institut National des Sciences Appliquées (INSA) Strasbourg, Strasbourg, France.
- Department of Surgery, University Hospital of Geneva, Geneva, Switzerland.
- Faculty of Medicine, Hebrew University of Jerusalem, Jerusalem, Israel.
Abstract
Artificial Intelligence (AI) is rapidly transitioning from experimental research to daily medical practice, yet the medical community's understanding of these tools remains largely confined to visible 'front-end' applications with which the clinician can directly interact (such as decision support systems and conversational agents). This perspective overlooks the proliferation of 'back-end' AI, the algorithms that are embedded within devices, systems, and hospital infrastructure that silently reconstruct data while remaining invisible to clinicians. This paper presents a clinician-oriented framework that distinguishes between these two categories and proposes a clinically oriented guide for their evaluation. We performed a narrative review of AI applications in minimally invasive therapy and medical imaging to categorize systems into 'front-end' (interactive/visible) and 'back-end' (embedded/invisible) modalities. We synthesized evaluation metrics from computer science and engineering literature, selecting those relevant for clinical safety and decision-making to create a practical literacy guide. Front-end and back-end systems require distinct validation strategies to ensure safety. while front-end evaluation must prioritize decision quality, spatial precision, and human-computer interaction to mitigate risks like automation bias, back-end evaluation requires rigorous technical benchmarking of signal fidelity and temporal latency to ensure that algorithmic reconstruction does not distort clinical reality. To facilitate this, we developed a structured inquiry framework to guide clinicians in auditing these systems for data provenance, transparency, and failure modes. Crucially, we emphasize that mathematical optimization does not guarantee clinical efficacy; technical metrics must always be paired with specific clinical contexts to ensure they align with patient-centered outcomes. Clinical safety in the AI era demands 'algorithmic literacy'. By applying this front-end/back-end framework and understanding key technical metrics, medical professionals can better identify failure modes, ensure data integrity, and maintain clear lines of clinical accountability, shifting from passive consumers to active evaluators of medical technology.