AI-Assisted Chromosome Aberration Analysis for Rapid Biological Dose Estimation of Ionizing Radiation

Radiation dose assessment in exposed individuals relies on the dicentric assay, the gold-standard cytogenetic biodosimeter that quantifies radiation-induced chromosomal aberrations. Although fully automated methods have been proposed, their accuracy remains suboptimal for large-scale emergency response. Here, we present a hybrid, semi-automatic framework that couples a deep-learning-based dicentric classifier with expert review. Among 2 000 metaphase images analyzed with a semi-automatic method, the specificity was 99.8%, the accuracy 98.7%, and the area under the curve (AUC) 0.977. For radiation dose estimation, the system achieved a minimal deviation of 9.25% and a maximal deviation of 15%. By integrating rapid AI triage with targeted human confirmation, the platform improves the high throughput required for population-scale triage while preserving the diagnostic rigor demanded by retrospective dose reconstruction.