Overview of imageCLEF 2018 medical domain visual question answering task

This paper presents an overview of the inaugural edition of the ImageCLEF 2018 Medical Domain Visual Question Answering (VQA-Med) task. Inspired by the recent success of visual question an-swering in the general domain, a pilot task was proposed this year to focus on visual question answering in the medical domain. Given medi-cal images accompanied with clinically relevant questions, participating systems were tasked with answering the questions based on the visual image content. A dataset of 6,413 question-answer pairs accompanied with 2,866 medical images extracted from PubMed Central articles was provided; from which, 5,413 question-answer pairs with 2,278 medical images were used for training, 500 question-answer pairs with 324 medi-cal images were used for validation, and 500 questions with 264 medical images were used for testing. Among 28 registered participants, 5 groups submitted a total of 17 runs, indicating a considerable interest in the VQA-Med task