In a study of 114 melanoma cases and 378 whole-slide images, a deep learning model found more mitotic hotspots than routine pathology counts, raising counts from 2.96 to 5.35 mitoses/mm2 (P = 0.004). The study also showed strong detection performance against expert review, with 88% sensitivity, 75% precision, and an F1 score of 0.81.

Why It Matters To Your Practice

• Mitotic activity is a key melanoma prognostic feature, but hotspot selection is prone to interobserver variability.

• AI support may help clinicians and pathologists identify the most proliferative tumor regions more consistently than routine review alone.

• Higher hotspot counts could affect risk stratification, pathology reporting, and downstream management discussions.

Clinical Implications

• The model detected 30,547 mitotic figures across the cohort, enabling whole-slide rather than limited-field assessment.

• AI-identified hotspots produced significantly higher counts than routine practice, suggesting current workflows may miss peak proliferative areas.

• If validated in practice, AI-assisted hotspot detection could serve as a second read for melanoma slides and support more standardized mitotic counting.

Insights

• Beyond counting mitoses, the algorithm characterized spatial distribution patterns ranging from clustered to relatively uniform across tumors.

• Melanomas from chronically sun-damaged sites had greater nearest-neighbour distances than non-CSD melanomas (423.3 μm vs. 285.4 μm, P = 0.023).

• That suggests AI pathology tools may eventually surface biologic patterns not routinely captured in standard reports.

The Bottom Line

• For clinicians interested in AI, this is a practical use case: better hotspot identification in melanoma, not just image classification.

• The near-term value is likely improved consistency and potentially more accurate prognostic assessment; the longer-term value is discovery of new spatial biomarkers.