Building a qualified annotation dataset for skin lesion analysis trough gamification

Abstract

The deep learning approach has increased the quality of automatic medical diagnoses at the cost of building qualified datasets to train and test such supervised machine learning methods. Image annotation is one of the main activity of dermatologists and the quality of annotation depends on the physician experience and on the number of studied cases: manual annotations are very useful to extract features like contours, intersections and shapes that can be used in the processes of lesion segmentation and classification made by automatic agents. This paper proposes the design of an interactive multimedia platform that enhance the annotation process of medical images, in the domain of dermatology, adopting gamification and "games with a purpose" (GWAP) strategies in order to improve the engagement and the production of qualified datasets also fostering their sharing and practical evaluation. A special attention is given to the design choices, theories and assumptions as well as the implementation and technological details.