A team from AUEB's Department of Informatics was ranked at 1st position in the ImageCLEFmed Caption 2020

A team from AUEB's Department of Informatics was ranked at positions 1, 2 and 6 among 49 systems that participated in the international ImageCLEFmed Caption 2020 competition (https://www.imageclef.org/2020/medical/caption/). Systems participating in the competition aim to automatically link medical images (e.g., X-rays) to medical concepts (e.g., body organs or malignancies appearing in the images). Systems of this kind can help retrieve medical images by keywords, but they can also be used as a first processing stage in larger systems that automatically generate parts of medical reports from medical images (http://nlp.cs.aueb.gr/pubs/sivl2019_survey_biomedical_image_captioning.pdf).

The team consisted of Vasilis Karatzas (undergraduate student of the Department of Informatics), Vasiliki Kougia (graduate of AUEB's MSc in Computer Science, http://grad.cs.aueb.gr/), John (Ioannis) Pavlopoulos (post-doctoral researcher of AUEB's Department of Informatics and Stockholm University, co-supervisor), and Ion Androutsopoulos (Professor of the Department of Informatics, co-supervisor), all members of AUEB's Natural Language Processing Group (http://nlp.cs.aueb.gr/). The team used systems based on deep neural networks to encode and classify the images, and information retrieval methods to retrieve similar images and concepts from past diagnoses. The systems were developed mostly during the BSc thesis project of Vasilis Karatzas, extending previous systems.

AUEB had also participated in last year's ImageCLEFmed Caption 2019, ranking at positions 1, 2, 3, 5 among approximately 60 systems (https://www.imageclef.org/2019/medical/caption/). Last year's team consisted of Vasiliki Kougia, John Pavlopoulos (co-supervisor), and Ion Androutsopoulos (co-supervisor). The systems of last year's team had been developed during the MSc thesis project of Vasiliki Kougia (http://nlp.cs.aueb.gr/theses/kougia_msc_thesis.pdf) and are described in an article in the proceedings of the CLEF 2019 conference (http://nlp.cs.aueb.gr/pubs/paper_136.pdf).

ID: 
2361