AI-based evaluation of testicular damage in stray dogs from environmentally polluted areas of the Campania Region, Italy

In Campania Region, an area known as the ‘Land of Fires’, has been described as an open-air dump due to the illegal discharges of toxic substances, raising concerns for animal and human health. Within a One Health perspective, stray dogs may act as sentinel species for environmental risk. For this purpose, testicular and epididymal samples from male stray dogs were classified according to spermatogenic status, ranging from normal spermatogenesis to intratubular seminoma. Heavy metal concentrations (cadmium, nickel, uranium) were quantified by inductively coupled plasma mass spectrometry. Tissue remodelling was assessed by histochemistry, immunofluorescence and Western blotting analyses for steroidogenic enzymes and cytoskeletal markers. Oxidative status was evaluated by H₂O₂ production and superoxide dismutase (SOD) activity. In parallel, H&E-stained digitized whole-slide images were analysed using an AI-based workflow with EfficientNet-B4 feature extraction and ensemble machine-learning classification, while epididymal sperm chromatin condensation was quantified on Toluidine Blue-stained cytological preparations through an automated k-means clustering pipeline. Our results highlighted that degenerated and neoplastic testes showed significantly higher concentrations of cadmium, nickel, and uranium compared to normal tissue. These changes were associated with progressive collagen and cytoskeletal disorganization, impaired steroidogenesis, and redox imbalance (increased H₂O₂, reduced SOD activity), alongside reduced epididymal sperm concentration and defective chromatin condensation. The AI ensemble classifier achieved 88.39% accuracy and a weighted F1-score of 0.79, with AUC values ≥0.92 across all histological categories. The Toluidine Blue automated pipeline showed high reproducibility (silhouette coefficient: 0.60 ± 0.09; adjusted Rand index: 0.89 ± 0.03), supporting the detection of impaired sperm chromatin condensation in degenerated cases. Overall, these findings support the potential value of dogs as sentinel species in One Health environmental risk assessment and suggest that AI-assisted histological and cytological analysis may strengthen the reproducibility of biomonitoring approaches in reproductive pathology.