The City PhD student wins the Medical Out-of-Distribution (MOOD) Analysis Challenge.

By Mr John Stevenson (Senior Communications Officer), Published

Sergio Naval Marimont, studying for a PhD in City’s Artificial Intelligence Research Centre (CitAI), has won First Prize in the Medical Out-of-Distribution (MOOD) research challenge for medical image analysis design.

His work was submitted to the 25th MICCAI Conference, held this year from September 18th to 22nd in Singapore.

706198Sergio, whose doctoral studies are supervised by Senior Lecturer, Dr Giacomo Tarroni, explains that “deep learning methods are increasingly leveraged to automatically analyse medical images.”

He says these methods rely on vast amounts of annotated images to identify patterns relevant to the medical image analysis tasks, commonly, the identification of pathologies.

Identifying anomalies

“In medical imaging, annotations are generated by radiologists who manually highlight anomalies or areas of interest that such methods need to identify. The cost of annotations and the availability of annotated images are factors limiting the clinical applications of deep learning methods. In a related limitation, traditional deep learning methods cannot be expected to identify anomalies
that are different from the ones present in the annotated datasets used for their training.”

“These constraints and limitations have led to the research question of how to leverage a large corpus of non-annotated images to train automated medical image analysis methods. Out-of-distribution methods address this research question by learning what comprises a healthy or in-distribution anatomy, leveraging only unannotated images from healthy subjects. Out-of-distribution methods are able to identify anomalies by contrasting new images to the patterns learnt from healthy images only.”

The prestigious annual MICCAI - International Conference on Medical Image Computing and Computer Assisted Intervention - attracts the world’s leading biomedical scientists, engineers, and clinicians from a wide range of disciplines associated with medical imaging and computer assisted intervention.

It includes three days of scientific content, including oral presentations and poster sessions. Workshops, tutorials, and challenges are held on the days preceding and succeeding the conference.

Led by Dr Eduardo AlonsoCitAI operates at the intersection between the development of novel AI techniques, Explainable AI (XAI) and Artificial General Intelligence (AGI), with a keen interest in the legal, ethical and social impact of AI.