.png)
Melanoma, an aggressive form of skin cancer, poses significant health risks to many individuals annually. Accurate diagnosis and prognosis of melanoma in its early stages are essential for determining the survival prospects of the patients and providing optimal treatment. Central to this assessment are key prognostic indicators: Breslow Thickness, the depth of the invasive tumor, and Ulceration, which denotes the breakdown of the skin atop the tumor. Evaluating such factors on pathology images is a laborious and time-consuming task for pathologists, potentially leading to possible inaccuracies in prognosis as well as prolonged wait times for patients awaiting tests and results. Hence, there is a pressing need for an automated system that can accelerate this assessment while ensuring accuracy. To the best of our knowledge, no previous studies in the literature have proposed an AI-based system specifically designed for estimating Breslow thickness and detecting Ulceration on melanoma pathology images.
In this study, we utilized two comprehensive datasets and investigated two learning methods using three cutting edge deep learning architectures. Our first method employs segmentation models, U-Net and SegFormer, trained with pixel-wise annotations. Our second method utilizes ConvNeXt, which learns from image-level labels. Our aim here was to develop and compare these models, assessing their performance in predicting Breslow thickness and identifying the presence of ulceration in Melanoma H&E-stained whole slide images.
Our experiments indicate that ConvNeXt outperforms its counterparts, demonstrating superior per- formance in estimating Breslow thickness, with a Mean Absolute Error of 0.54 mm, and detecting Ulceration, achieving an accuracy of 88%. A pivotal advantage of ConvNeXt is its capacity to learn from global labels, eliminating the need for pixel-wise annotations, thus highlighting its scalability and cost-effectiveness.
The results of this study underscore the potential of ConvNext as a promising tool in melanoma prognosis and treatment planning. Its superior performance, combined with its practical advantages in terms of training requirements, suggests that it could serve as a valuable asset in the fight against melanoma, subject to further validation and integration into clinical workflows. Furthermore, we believe that this study lays a foundation for future AI systems to prognosticate not just melanoma severity but also potential patient survival duration and cancer recurrence probabilities.
Y Salhi, J Rynkiewicz, C Bossard, A Nakhjavani, S Salhi, J Chetritt