Micro- and nanoplastics in food and water from farm-to-fork: exposure pathways, human biomonitoring and AI-enhanced detection – a systematic review

Micro- and nanoplastics (MNPs) are now globally recognised as emerging contaminants across food and water systems, but inconsistent and non-standardised methodologies continue to hinder a clear understanding of human exposure. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, 55 studies were systematically reviewed across two tiers: exposure (farm-to-fork) and AI-based detection. This two-tiered systematic review provides the first integrated “farm-to-fork synthesis” linking environmental entry points, food processing and packaging, consumption and human biomonitoring evidence, alongside the application of artificial intelligence (AI) for MNP detection. Across 37 Tier One exposure studies, fibres dominated across multiple food chain stages, with concentrations highest in aquaculture and processed foods, while terrestrial crops and soils remain critically understudied. Human biomonitoring data confirm ingestion and systemic presence of MNPs in stool, urine, blood and tumour tissues, identifying diet as a major exposure route. Tier Two evaluated 18 studies applying AI to Raman spectroscopy and Fourier transform infrared spectroscopy (FTIR), where machine learning (ML) and deep learning (DL) models achieved expert-level accuracy, improved reproducibility and reduced annotation time, though dataset heterogeneity remains a limitation. Together, these findings highlight food-system stages most susceptible to MNP contamination, providing actionable priorities for targeted monitoring, mitigation and regulatory development. They also demonstrate the potential of AI-enhanced spectroscopy to harmonise MNP detection. However, reported accuracies above 95% largely derive from curated or synthetic spectral libraries and external validation on complex environmental, food and water matrices remains limited. To strengthen comparability across studies, we recommend minimum reporting standards including dataset provenance, spectrum pre-processing, validation strategies, polymer classes, colourants and misclassification reporting, which could enable standardised biomonitoring and inform food safety policy.