Machine learning and optical imaging for pharmaceutical forensic toxicology: A comprehensive review

Abstract

Pharmaceutical forensic toxicology is undergoing a profound transformation driven by the convergence of optical imaging technologies and machine learning methodologies. Traditionally focused on post hoc legal investigations, the field is increasingly expanding toward proactive roles in pharmaceutical quality control, counterfeit drug detection, and public health protection. This review provides a comprehensive and critical overview of the integration of machine learning–based chemometrics with optical imaging techniques in pharmaceutical forensic toxicology. Imaging modalities ranging from grayscale and red, green, blue (RGB) imaging to infrared, Raman, multispectral, and hyperspectral imaging (MSI & HSI) are discussed, with emphasis on their physical principles, data structures, and analytical capabilities. The role of supervised and unsupervised chemometric multivariate models is examined in the context of classification, authentication, and quantitative assessment of pharmaceutical products. Current applications are reviewed across key forensic scenarios, including optical identification of counterfeit and illicit drugs, non-destructive evaluation of confiscated products, retrospective toxicological investigations, and emerging portable artificial intelligence-enabled platforms. Beyond technical performance, this review critically addresses regulatory, ethical, and legal challenges associated with artificial intelligence in forensic environments, highlighting the importance of explainability, traceability, and data governance. Finally, future perspectives are discussed, emphasizing the transition toward integrated forensic ecosystems that combine optical imaging, spectral databases, and interpretable machine learning to support robust, transparent, and legally defensible toxicological decision-making.