Multimodal Multi-tasking for Skin Lesion Classification Using Deep Neural Networks

Abstract Skin cancer is one of the most common types and, with its increasing incidence, accurate early diagnosis crucial to improve prognosis patients. In process visual inspection, dermatologists follow specific dermoscopic algorithms and identify important features provide a diagnosis. This can be automated as such characteristics extracted by computer vision techniques. Although deep neural networks extract useful from digital images for skin lesion classification, performance improved providing additional information. The pseudo-features used input (multimodal) or output (multi-tasking) train robust learning model. work investigates multimodal multi-tasking techniques more efficient training, given single optimization several related tasks in latter, generation better predictions. Additionally, role segmentation also studied. Results show that improves beneficial which lead predictions, inspired ABCD rule readily available helpful information about lesion.