Project Title: Skin Lesion Classification Using Deep Learning
Area of Research: Computer Vision (CV)
Problem Statement:
The project aims to develop a deep learning model for accuratelv classifving skin lesions into distinct categories. including melanoma, nevi, and benign lesions, Skin lesion classification is a critical task in dermatology, and automatinn thie nrocoee ueinn deen learninn tochniauee can aid in oarlv dotection and imnroved nationt outcomoe Unlike standard datasets such as MNIST or CIFAR-10, which are limited in diversity, skin lesion classification presents challenges due to the variability and complexity of skin conditions.
Dataset:
Figure 1: example images of ISIC dataset.
The project utilizes the International Skin Imaging Collaboration (ISIC) dataset, which contains a diverse collection ofl dermatoscopic images encompassing various skin lesion tvpes. The dataset is annotated with ground truth labels for lesion categories, providing a valuable resource for supervised learning tasks in skin lesion classification.
Dataset URL: https://challenge.isic-archive.com/data/
Task:
The task involves training a deep neural network to accurately classify skin lesions into predefined categories.l including melanoma (malignant), nevi, and benign lesions, Students will explore architectures such as ResNet, VGG, or custom networks suitable for handling complex image features specific to dermatoscopic images. Data augmentation techniques such as rotation, flipping, scaling, and color jittering will be employed to improve model generalization and robustness.
Relevant Papers:
1. Kassem, Mohamed A., et al. "Machine learning and deep learning methods for skin lesion classification andl diagnosis: a systematic review." Diagnostics 11.8 (2021): 1390.
2. Lopez. Adria Romero, et al. "Skin lesion classification from dermoscopic images using deep learning techniques." 2017 13th IASTED international conference on biomedical engineering (BioMed). IEEE, 2017.
3. Benyahia, Samia, Boudjelal Meftah, and Olivier Lézoray. "Multi-features extraction based on deep learning for skinl lesion classification."Tissue and Cell 74 (2022): 101701.