Towards Accurate Breslow Measurements: Mitigating Issues in Histopathological Imaging.

Breslow thickness is a key prognostic parameter in the staging of cutaneous melanoma, but its manual measurement is affected by operator dependency and the complex morphology of the epidermis. Identifying both the deepest melanocyte and the correct perpendicular path to the epidermal surface can be challenging, especially in highly irregular tissues. This study investigates a more robust estimation of Breslow thickness through the development of a semi-automated Computer Vision-based software. Inter-operator variability was assessed by comparing measurements performed by seven histopathologists on 40 Whole Slide Images of non-ulcerated pT1a melanomas with Breslow thickness below 0.8 mm. The agreement between human measurements and AI results was evaluated. Significant variability in the orientation of the measurement segments was observed, highlighting the difficulty for human operators in identifying the correct perpendicular direction. A linear relationship was found between angular variance and variance in Breslow thickness values, linking epidermal irregularity to increased measurement uncertainty. Overall, statistically significant differences were observed between the AI system and five of the seven operators, indicating a general tendency among experts to overestimate Breslow thickness.