Artificial intelligence for medication recognition and safety: a systematic review of detection, classification, OCR, and digital health applications
- 1School of Computer Science Engineering and Information Systems, Vellore Institute of Technology , , ,
Abstract
Introduction
Medication errors and adverse drug events remain causes of patient harm, underscoring the need for technologies supporting medication identification, verification, and medication safety. Artificial intelligence (AI), computer vision, optical character recognition (OCR), and digital health technologies are applied to medication recognition.Objective
To systematically review AI-based medication recognition systems and characterize recognition tasks, computational approaches, datasets, evaluation, validation, generalizability, deployment, applications, and emerging directions.Methods
A systematic review was conducted in accordance with PRISMA 2020. Scopus, Web of Science, IEEE Xplore, and PubMed/MEDLINE were searched for publications from 1 January 2016 to 4 September 2026, using two conceptual blocks covering medication recognition and AI methods. Of 895 retrieved records, 136 duplicates were removed, and 759 records underwent screening. After full-text assessment, 14 reports were excluded, and 91 studies were included. Title-and-abstract screening yielded 84.72% observed agreement (Cohen's κ=0.527 ). For methodological assessment, both reviewers independently applied D1-D8 to the 91 included studies before consensus resolution. Across 728 domain-level assessments, observed agreement was 97.66%, with an aggregate Cohen's κ of 0.959; domain-specific κ values ranged from 0.857 to 1.000. Evidence was synthesized through cross-study analysis and four evidence maps.Results
Evidence shows use of detection, classification, OCR/imprint recognition, hybrid and multimodal architectures, alongside few-shot, low-shot, zero-shot, and incremental learning. Applications extend to assistive, mobile, and IoT-enabled systems, medication verification, and clinical decision support. Heterogeneity in datasets, imaging conditions, evaluation protocols, and validation strategies limits comparison and generalizability. External validation, cross-dataset testing, unseen-medication evaluation, prospective deployment, and clinical testing remain limited.Conclusion
AI-based medication recognition is evolving from a computer vision task into a digital informatics component supporting medication safety, accessibility, verification, and patient care. Future research should prioritize diverse datasets, leakage control, independent multicenter validation, unseen-medication evaluation, explainable and privacy-preserving AI, interoperability, and resource-efficient deployment. Evaluation should extend beyond accuracy to clinical utility, safety, usability, trust, and patient-centered outcomes.
Related
- Artificial Intelligence and Anatomical Variation: A Narrative Review of Detection, Classification, and Educational Applications.
- Hidden Signals in Serum Protein Electrophoresis: Artificial Intelligence from Laboratory Pattern Recognition to Clinical Validation
- Attitudes of people living with dementia and their carers towards the use of generative artificial intelligence to inform structured medication reviews.
- From Pattern Recognition to Decision Guidance: Artificial Intelligence in Interstitial Lung Disease Diagnosis and Surveillance
- Artificial Intelligence-Driven Nanobody Engineering for Pathogen Recognition: From Functional Epitope Prioritization to Multimodal Foundation Models
BibTeX
@article{sachin2026artificial,
title = {{Artificial intelligence for medication recognition and safety: a systematic review of detection, classification, OCR, and digital health applications}},
author = {A. Sai Sachin and J. Karthikeyan},
journal = {Frontiers in digital health},
year = {2026},
month = oct,
doi = {10.3389/fdgth.2026.1963607},
url = {https://europepmc.org/article/PMC/PMC13717275},
note = {CC BY},
}
RIS
TY - JOUR
TI - Artificial intelligence for medication recognition and safety: a systematic review of detection, classification, OCR, and digital health applications
AU - A. Sai Sachin
AU - J. Karthikeyan
PY - 2026
DA - 2026/10/15
JO - Frontiers in digital health
DO - 10.3389/fdgth.2026.1963607
UR - https://europepmc.org/article/PMC/PMC13717275
AB - Introduction Medication errors and adverse drug events remain causes of patient harm, underscoring the need for technologies supporting medication identification, verification, and medication safety. Artificial intelligence (AI), computer vision, optical character recognition (OCR), and digital health technologies are applied to medication recognition.Objective To systematically review AI-based medication recognition systems and characterize recognition tasks, computational approaches, datasets, evaluation, validation, generalizability, deployment, applications, and emerging directions.Methods A systematic review was conducted in accordance with PRISMA 2020. Scopus, Web of Science, IEEE Xplore, and PubMed/MEDLINE were searched for publications from 1 January 2016 to 4 September 2026, using two conceptual blocks covering medication recognition and AI methods. Of 895 retrieved records, 136 duplicates were removed, and 759 records underwent screening. After full-text assessment, 14 reports were excluded, and 91 studies were included. Title-and-abstract screening yielded 84.72% observed agreement (Cohen's κ=0.527 ). For methodological assessment, both reviewers independently applied D1-D8 to the 91 included studies before consensus resolution. Across 728 domain-level assessments, observed agreement was 97.66%, with an aggregate Cohen's κ of 0.959; domain-specific κ values ranged from 0.857 to 1.000. Evidence was synthesized through cross-study analysis and four evidence maps.Results Evidence shows use of detection, classification, OCR/imprint recognition, hybrid and multimodal architectures, alongside few-shot, low-shot, zero-shot, and incremental learning. Applications extend to assistive, mobile, and IoT-enabled systems, medication verification, and clinical decision support. Heterogeneity in datasets, imaging conditions, evaluation protocols, and validation strategies limits comparison and generalizability. External validation, cross-dataset testing, unseen-medication evaluation, prospective deployment, and clinical testing remain limited.Conclusion AI-based medication recognition is evolving from a computer vision task into a digital informatics component supporting medication safety, accessibility, verification, and patient care. Future research should prioritize diverse datasets, leakage control, independent multicenter validation, unseen-medication evaluation, explainable and privacy-preserving AI, interoperability, and resource-efficient deployment. Evaluation should extend beyond accuracy to clinical utility, safety, usability, trust, and patient-centered outcomes.
SN - 2673-253X
ER -