Beyond high accuracy: a cross-technology framework for validating AI driven mycotoxin screening in complex food matrices

Guo, Z., Feng, L., Li, W., Xing, K., Guo, X., Cao, L. and Auat Cheein, F. (2026) Beyond high accuracy: a cross-technology framework for validating AI driven mycotoxin screening in complex food matrices. Trends in Food Science & Technology, 178. ISSN 09242244

Full text not available from this repository.

Abstract

Abstract Background Recent literature frequently couples optical sensing with machine learning for mycotoxin screening, often reporting accuracies exceeding 95%. However, industrial adoption appears limited. We hypothesize this translational gap likely stems from a systemic disconnect between laboratory-generated computational accuracy and the physical robustness required in field environments. Scope and approach Unlike previous reviews cataloging algorithmic advancements, this study takes a fundamentally different, problem-driven approach. Synthesizing primary literature (2020–2026), we found that exceptional performances can sometimes be influenced by methodological artifacts: conflating fungal biomass with direct toxin detection, bypassing natural infections via artificial spiking, and using internal validation susceptible to spatial data leakage. To systematically re-evaluate published models against deployment criteria, the evidence robustness framework (ERF) was introduced, assessing sample authenticity, validation rigor, matrix dynamics, regulatory alignment, and model transparency. Key findings and conclusions To enhance comparability and regulatory legibility, the minimum information for AI modeling (MI-AIM) reporting checklist is proposed. Transitioning from laboratory to industrial deployment will likely require broader methodological shifts rather than isolated algorithmic enhancements. By outlining actionable pathways, including multimodal digital twins and physics-informed explainable AI, this review aims to guide a cultural shift toward transparent reporting of external-validation metrics, helping establish these platforms as trusted, industry-ready food-safety tools.

Item Type:	Article
Additional Information:	Full text not available from this repository. Copyright remains with © 2026 Elsevier Ltd. All rights reserved.
Keywords:	Mycotoxins, Hyperspectral imaging, Explainable AI, Evidence robustness, Food safety, Deep learning
Divisions:	Research Centres > Harper Institute of Technology
Depositing User:	Miss Anna Cope
Date Deposited:	14 Oct 2026 08:39
Last Modified:	14 Oct 2026 08:39
URI:	https://hau.repository.guildhe.ac.uk/id/eprint/18434

Actions (login required)

Edit Item