Deep learning and object detection methods for scoring cell types within the human buccal cell micronucleus and cytome assays for human biomonitoring

Micronuclei (MN) are critical biomarkers for pathological conditions, yet their manual scoring is inherently laborious and prone to significant interobserver variability, limiting the reliability and scalability of genotoxicity assessments. Recent advancements in deep learning and computer vision have revolutionized automated MN detection in various assay samples, enhancing accuracy and efficiency and reducing human bias. While these artificial intelligence (AI)-powered techniques have been demonstrated in in vitro genotoxicity testing, their application to the minimally invasive buccal micronucleus cytome (BMCyt) assay for human biomonitoring remains largely unexplored. The BMCyt assay, invaluable for assessing genotoxic damage in environmentally exposed populations, presents unique challenges, including sample variability, confounding factors, and the complexity of scoring multiple cytogenetic endpoints. This review covers the evolution of AI-based MN detection, analysing key methodologies and advancements. It highlights the untapped potential of integrating AI into the BMCyt assay to overcome current analytical limitations, improve reproducibility, increase throughput, and eliminate observer bias. By facilitating more robust and scalable genomic damage monitoring, AI integration will significantly enhance the utility of the BMCyt assay in large-scale epidemiological studies and human biomonitoring.