Authors
Gianmarco Sirago, Susanna Sabato, Paolo Visci, Biagio Solarino, Alessandro Dell'Erba, Davide Ferorelli
Published in
Forensic science international. Volume 388. Pages 113110. Aug 18, 2026. Epub Aug 18, 2026.
Abstract
Artificial intelligence (AI) is rapidly entering the forensic sciences. We mapped AI research produced within dedicated forensic-science journals to characterise its growth, contributors, intellectual structure and emerging themes, and to assess whether methodological, legal and security-critical issues have developed alongside applications.
We analysed documents published 1991-2026 in dedicated forensic-science journals (Scopus category "Pathology and Forensic Medicine", applied by ISSN), retrieved from Scopus and the Web of Science and merged in bibliometrix; by design the study maps AI research produced within forensic-science journals rather than all forensically relevant AI. Analyses covered production, sources, countries, keyword co-occurrence, thematic mapping, trend topics, the intellectual base, and a targeted document-level analysis of fifteen methodological, legal and security domains.
The corpus comprised 1127 documents from 42 sources and 4123 indexed author name-forms, with a 16.29% annual growth rate and mean document age of 3.76 years. Output was negligible before 2015 and rose steeply thereafter. Forensic Science International was the most productive source, followed by the International Journal of Legal Medicine; China, India and the United States led corresponding-author output, with low international co-authorship among the largest producers. The conceptual structure resolved into a machine-learning/AI/identification core, a deep-learning/imaging/age-estimation component and a forensic-anthropology theme. Whereas generic performance evaluation was common (≈59% of documents), the terminology of critical forensic-qualification themes was uncommon - these are counts of records in which the vocabulary is visible in title, abstract or keywords, not of studies verified to have performed the procedure: external/independent validation 2.8%, calibration or predictive-uncertainty language 0.4%, admissibility 4.2%, adversarial/model-security 0.8%, bias/fairness 1.9% and dedicated law/regulation/AI-Act content 4.0% (with AI applied to cybercrime, a comparative category rather than a safeguard, at 3.5%).
AI within dedicated forensic-science journals is a young, rapidly expanding and geographically concentrated field consolidated around machine and deep learning applied to identification, anthropology and age estimation. The limited visibility of external validation, adversarial resilience and evidentiary standards defines a practical agenda for multicentre datasets, transparent reporting and forensic-specific validation within that scope.
PMID:
42617285
Bibliographic data and abstract were imported from PubMed on 20 Aug 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 4
- Comments 0