Detection and Segmentation of Malignant Melanoma Regions in Dermoscopic Images Using Machine Learning

Melanomas are among the deadliest types of skin cancer, requiring early detection to improve survival rates. This project employs deep learning techniques, specifically U-Net-based architectures, to segment and identify melanomas in an open dataset of dermoscopic images. For the segmentation task, the proposed models achieved a highest Dice score of 0.88 and an Intersection over Union (IoU) of 0.80, demonstrating their effectiveness in delineating lesion boundaries. In the classification task, the approaches ranged from the use of Convolutional Neural Networks (CNNs) to embedding extraction combined with machine learning models. The bestperforming classification model attained an overall accuracy of 0.84, with an F1-score of 0.45 for the melanoma class. These results highlight the potential of the developed methodologies to enhance diagnostic precision and support dermatologists in identifying high-risk cases more effectively.

• Autores: Martínez Novoa, Santiago
• Tipo de recurso: Trabajo de grado de pregrado
• Fecha de publicación: 2025
• Institución: Universidad de los Andes
• Repositorio: Séneca: repositorio Uniandes
• Idioma: eng
• OAI Identifier: oai:repositorio.uniandes.edu.co:1992/75910
• Acceso en línea: https://hdl.handle.net/1992/75910
• Palabra clave: Deep learning; Melanoma detection; U-Net; Segmentation; Convolutional neural networks; Ingeniería
• Rights: openAccess
• License: Attribution 4.0 International