Abstract / Summary
Artificial intelligence (AI) is being investigated as a means to advance forensic histopathology from visual description to quantitative mapping, pattern recognition, and multimodal reconstruction of death. Recent developments in human postmortem analysis include validated whole-slide digitisation as foundational infrastructure and task-specific AI applications for diatom detection, glomerular quantification, coronary morphometry, myocardial classification, immunohistochemistry scoring, and virtual staining. AI-based postmortem interval (PMI) estimation remains experimental; the multi-organ study referenced in this commentary utilised an animal model, and human forensic validity has not yet been established. Foundation and vision-language models may ultimately extend these capabilities across organs and integrate microscopic findings with autopsy, toxicological, radiological, and circumstantial evidence. Postmortem tissue, however, is subject to autolysis, decomposition, agonal events, resuscitation, and non-standardised sampling, while the reference diagnosis is often a probabilistic whole-case opinion rather than an objective slide label. This work introduces FORENS-AI (FORensic Histopathology Evidence Network and Safeguards for AI), a conceptual architecture designed to convert computational outputs into traceable medicolegal evidence. FORENS-AI has not been empirically validated and is not presented as a standard, scoring system, clinical decision rule, or practice recommendation. The framework represents findings as an evidence graph, separates the Slide, Case, and Claim Gates, and proposes progressively stronger validation as outputs move from measurement to temporal, mechanistic, or causal inference. The primary aim is not to automate certification of death, but to identify safeguards to evaluate before AI-derived histological findings are permitted to inform expert opinion.