SKIN LESION CLASSIFICATION USING CONVOLUTIONAL NEURAL NETWORKS

Skin cancer is a common cancer with a large count of patients diagnosed annually. Early diagnosis and correct classification of skin lesions may increase the likelihood of cure before it turns malignant and the cancer metastasises. In this study we have trained the system using the International Skin Imaging Collaboration (ISIC) curated datasets of a huge number of gold-standard lesion diagnosed training images from patients with different skin disease conditions. Convolutional neural networks are used in this study because of their superior performance in medical imaging. Various architecture models and data augmentation strategies are investigated to alleviate dataset imbalances and improve model resilience. The comparative study of model performance demonstrates the superiority of InceptionV3 model in terms of validation accuracy and computational efficiency.