Context-Aware Data Augmentation for Detection, Segmentation, and Quantitative Analysis of Mitotic Figures in Melanoma Whole-Slide Images.

Mitotic figure counting is critical for evaluating melanoma proliferation and prognosis, but manual counting is subjective and poorly reproducible, and the sparse distribution of mitotic figures in whole-slide images leads to a scarcity of training data. To address this, we propose a context-aware copy-paste data augmentation framework (CACPASTE) that integrates histopathological priors. CACPASTE ensures biological rationality and morphological fidelity through tissue region constraints and cell morphology simulation, and uses an adaptive fusion strategy for natural background integration. In U‑Net segmentation experiments on an independent test set, the proposed method achieves a Dice coefficient of 0.7861, outperforming traditional augmentation and performing comparably to InstaBoost and SPADE. In a segmentation‑guided detection cascade (U‑Net‑YOLOv8), we obtain a precision of 0.9023, recall of 0.9373, and F1‑score of 0.9231, simultaneously outputting segmentation masks and detection boxes to unify both tasks. Furthermore, the system automatically extracts and quantifies nuclear morphological indicators (area, circularity, aspect ratio) and spatial distribution characteristics (density, clustering) of mitotic figures. This multi‑dimensional quantitative evaluation provides objective, reproducible computational pathology tools for precise diagnosis and prognosis assessment of melanoma.