A novel nomenclature for the identification of ground truth in medical imaging data : Design, implementation and integration in a large knowledge database

Abstract: The annotation of medical images is a critical task for many downstream applications. However, the lack of a unified annotation nomenclature has resulted in inconsistency and ambiguity in the storage and use of such data. In this thesis, we propose and evaluate a novel annotation nomenclature for medical images. Our nomenclature is designed to be intuitive, easy to use and to expand. We also developed a knowledge database storing large medical image datasets that integrates the new nomenclature. The database is implemented as a server application exposing REST APIs. This allows users to upload/download datasets and query the data based on the annotations and to integrate the system in existing frameworks. We conducted a user study to assess the usability characteristics of the label nomenclature and its integration in the new system. The results collected from the user base are positive. The nomenclature is well perceived and the users had rated positively the usability of the whole system.