Identification and Localization of Malanoma Skin Cancer in Neuropathological Images

Early detection and treatment of melanoma can improve survival rates. The demand for computational pathology (CPATH) systems has been emphasized by a predicted rise in skin cancer cases and a shortage of dermatopathologists. Deep learning (DL) models in CPATH systems have the capability of using underlying morphological and cellular cues to detect the existence of melanoma. This study suggests a new approach to identify melanoma and distinguish benign from malignant melanocytic tumors in whole slide images (WSI). With excellent accuracy, this technique locates lesions on a WSI to highlight possible areas of interest for pathologists. It's interesting to note that our DL technique uses a single CNN network to build localization maps first, then uses those maps to make slide level predictions to identify patients with melanoma. With a 0.992 F1 score and 0.99 sensitivity on unseen data, our top model delivers excellent patch-wise classification results