Skin Cancer Classification Using Machine Learning and Convolutional Neural Networks

Abstract

The timely and accurate diagnosis and treatment of skin cancer are crucial as it is a common and potentially deadly condition. One effective machine learning technique for accurately classifying skin lesions and improving skin cancer diagnosis is known as convolutional neural networks (CNNs). The article offers a detailed examination of the machine learning and CNN techniques employed for skin cancer classification. The discussion includes several studies that have utilized these methods to classify various types of skin lesions, such as basal cell carcinoma, melanoma, and squamous cell carcinoma. Nevertheless, the lack of high-quality datasets presents a significant challenge in skin cancer classification. To address this problem, one of the large-scale datasets developed is the ISIC dataset, which features over 20,000 annotated images of skin lesions. Skin cancer classification has been tackled using different machine learning algorithms, such as decision trees, support vector machines, random forests, as well as CNNs. CNNs are particularly useful for accurately classifying skin lesions because they can learn complex features from images. Several studies have utilized pre-trained CNNs, such as VGGNet and Inception, to classify skin lesions, achieving high accuracy rates of 85–95%. Ensemble methods, such as bagging and boosting, have also been used to improve skin cancer classification accuracy. These methods combine the predictions of multiple models to achieve better performance than a single model. Transfer learning has been effective in reducing overfitting and improving model generalization performance. This review also highlights studies that have utilized multi-modal data, such as dermoscopy and clinical images, to improve skin cancer classification accuracy by providing complementary information about skin lesions.