LRNet: Skin Cancer Classification using Low-Resolution Images

Modern lifestyle and continuous depletion of the ozone layer have dramatically increased skin cancer cases worldwide. Detection of skin cancer at an early stage can increase the survival rate to 99%. Computer-aided skin cancer classification has various challenges due to the visual similarity between different skin cancer classes and model’s lack of generalization capability. Recent deep learning-based approaches have overcome these challenges and even outperform an expert dermatologist. However, these models are trained on a large number of high-resolution images which makes them computationally expensive. To elucidate this problem, we propose to train the neural networks on low-resolution images. We then propose a fast, scalable, and efficient Deep Convolutional Neural Network, LRNet for skin cancer classification on low-resolution images. LRNet is trained on 10015 pigmented lesion images belonging to the HAM10000 dataset. The proposed model is also deployed to a publicly available web application. The performance of LRNet is evaluated based on sensitivity, specificity, precision and accuracy, and the values obtained are 94%, 91.7%, 94.2%, and 90.6%, respectively. Our proposed approach outperforms all the pre-trained models and drastically reduces the training time.