Early Detection of Melanoma Through Deep Learning-Based Skin Lesion Classification Using VGG16 and InceptionV3

This work aims at investigating classification of skin lesion in order to detect melanoma using deep learning models such as VGG16 and InceptionV3. Skin cancer especially the melanoma needs proper identification for cure to be administered on the affected tissues. The research which underpins this application makes use of convolutional neural networks (CNNs) in an effort to decide whether skin lesions are benign or malignant. In the present study, to assess the performance of the suggested models, the dermoscopic image data set in the public domain was used for training VGG16 and InceptionV3 models. The experimental evaluation demonstrated that, in comparison to the VGG16 network and other relevant algorithms, including ResNet50 and DenseNet121, InceptionV3 provides higher accuracy, precision, recall, F1-score, as well as the ROC-AUC value. InceptionV3 delivered 91.5 % classification accuracy and 92.1 % F1 score representing the enhanced melanoma diagnosis capability of the algorithm with fewer statistical errors between the true positive and true negative cases. The broad fractal dimension analysis feature of the proposed model means that variations in lesion size and texture do not pose a challenge to it. The deep learning models take more time in training than the traditional machine learning methods, although their outcomes justify their employment in clinical practices. The study identifies InceptionV3 as a viable tool for analysis of melanoma images with a view to assist the doctors in early identification of the disease hence enhancing the generality of patients prognosis.