Distributed Learning for Melanoma Classification using Personal Health Train

• DOI: 10.48550/arxiv.2103.13226
• Publication Date: 2021-01-01
• Authors: Yongli Mou; Sascha Welten; Yeliz Üçer Yediel; Toralf Kirsten; Oya Beyan
• Keywords: FOS: Computer and information sciences; Computer Science - Distributed, Parallel, and Cluster Computing; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Distributed, Parallel, and Cluster Computing (cs.DC); 3. Good health

Skin cancer is the most common type. Usually, patients with suspicion of are treated by doctors without any aided visual inspection. At this point, dermoscopy has become a suitable tool to support physicians in their decision-making. However, clinicians need years expertise classify possibly malicious skin lesions correctly. Therefore, research applied image processing and analysis tools improve treatment process. In order perform train model on dermoscopic images data needs be centralized. Nevertheless, centralization does not often comply local protection regulations due its sensitive nature loss sovereignty if providers allow unlimited access data. A method circumvent all privacy-related challenges Distributed Analytics (DA) approaches, which bring instead vice versa. This paradigm shift enables analyses - our case, remaining inside institutional borders, i.e., origin. documentation, we describe straightforward use case including training for lesion classification based decentralised