Whole‐slide analysis of mitotic activity in melanoma reveals higher‐proliferation hotspots and distinct spatial patterns

Introduction Accurate assessment of mitotic activity is an important component of melanoma prognostication but remains challenging due to interobserver variability and difficulties in identifying representative mitotic hotspots. We developed and evaluated a deep learning–based mitosis detection algorithm for whole‐slide images (WSIs) of melanoma and investigated its utility for hotspot identification and spatial analysis of mitotic distribution. Methods The model was trained on manually annotated histopathology images and applied to a cohort of 114 melanoma cases comprising 378 WSIs. Performance was assessed against expert annotation using sensitivity, precision, and F1‐score. Algorithm‐identified hotspots were reviewed by a pathologist and compared with mitotic counts reported in routine clinical practice. Spatial organization of mitotic figures was evaluated using nearest‐neighbour distance analyses. Results The algorithm achieved a sensitivity of 88%, precision of 75%, and an F1 score of 0.81, demonstrating strong agreement with expert assessment. Across the cohort, 30,547 mitotic figures were detected, enabling comprehensive whole‐slide analysis of proliferative activity. AI‐assisted hotspot identification yielded significantly higher mitotic counts than those reported in routine pathology practice (5.35 ± 7.9 versus 2.96 ± 3.73 mitoses/mm 2 , P = 0.004), suggesting improved identification of regions with maximal proliferative activity. Whole‐slide analysis further enabled characterization of mitotic spatial organization. Distinct patterns ranging from clustered to relatively uniform distributions were observed across tumours. Moreover, melanomas arising in chronically sun‐damaged (CSD) sites demonstrated significantly greater nearest‐neighbour distances than melanomas from non‐CSD sites (423.3 μm versus 285.4 μm, P = 0.023), indicating differences in the spatial organization of proliferating tumour cells. Conclusion Our findings demonstrate that AI‐assisted mitosis detection can accurately identify mitotic figures while improving hotspot detection compared with routine assessment. Beyond mitotic quantification, large‐scale whole‐slide analysis enables novel spatial characterization of mitotic organization, providing additional insights into melanoma biology and highlighting new opportunities for computational pathology–based biomarker discovery.