<![CDATA[Cyberattacks threaten healthcare data security and clinical continuity. Although artificial intelligence (AI) promises improved intrusion detection, high accuracy rarely translates into clinical deployment. This systematic review examines evaluation practices and deployment conditions in healthcare intrusion detection research. Following PRISMA 2020, searches across ScienceDirect, Scopus, PubMed, MDPI and SpringerLink covered studies published between 2020 and 2024. Eleven studies met the inclusion criteria. Findings were triangulated against subsequent large-scale reviews to assess consistency across broader evidence. Classical machine learning dominated, with limited adoption of hybrid, federated or transformer-based approaches. Reported accuracy averaged 98.66%, predominantly using benchmark or synthetic datasets under controlled conditions. False positive rates, alert volumes, inference latency and robustness to distribution shift were rarely reported. No included study demonstrated sustained operation in a live clinical network. The review introduces the Healthcare IDS Deployment Readiness Assessment (HIDRA), comprising five dimensions and five ordinal readiness levels, alongside a minimum reporting set and six testable propositions. No study exceeded readiness level 2 of 5. HIDRA offers authors, reviewers and procurement teams a reusable instrument for assessing deployment readiness and identifying gaps between experimental performance and clinical feasibility.]]>
Copyrights © 2026
