15 October 2020
School of Biomedical Engineering & Imaging Sciences success at imaging conference MICCAI
The School had another successful year at the annual imaging conference
Researchers from the School of Biomedical Engineering & Imaging Sciences showcased their research excellence at this year’s International Conference on Medical Image Computing & Computer Assisted Intervention (MICCAI) held online.
Upwards of 50 representatives from the School participated in this year’s conference, including keynote presenters, event chairs, outstanding reviewers and multiple awardees for best papers and challenges.
Head of School Professor Sebastien Ourselin said the School’s success at MICCAI was indicative of the strength of the imaging research within the organisation.
“Each year our imaging colleagues from the School demonstrate their strong research and output at MICCAI. From students to Professors, these representatives engage with activities, deliver keynotes, organise sessions and push forward the School’s agenda,” he said.
School’s senior academics showcased strong leadership
Keynote speakers from the School included Professor Julia Schnabel, Professor Alistair Young, Dr Tom Booth, Dr Jorge Cardoso, Esther Puyol-Anton, and Alena Uus.
Professor Julia Schnabel, Director of the School’s EPSRC CDT in Smart Medical Imaging and MICCAI Society General Secretary gave an invited ‘Women in MICCAI’ talk and live online panel on ‘My career - a work in progress’.
Professor Schnabel also moderated the first ever MICCAI Town Hall meeting and gave atalk at the ‘Predictive Intelligence in Medicine (PRIME)’ MICCAI satellite workshop on ‘Predictive intelligence for image quality, detection and progression’.
"It was a pleasure to share my career journey with the audience at MICCAI and reflect on the advances women have made in the field," she said.
Satellite events were chaired by Dr Jana Hutter (Computational Diffusion MRI), Dr Andrew Melbourne with Dr Jana Hutter (Perinatal, Preterm and Pediatric Image Analysis), Dr Jorge Cardoso (Domain Adaptation and Representation Transfer, Distributed and Collaborative Learning), Professor Alistair Young (Statistical Atlases and Computational Modelling of the Heart) and Dr Emma Robinson (Machine Learning in Clinical Neuroimaging).
Dr Jana Hutter also organised the Computational Diffusion MRI challenge.
The aim of the Super-resolution of Multi-dimensional Diffusion MRI (Super MUDI) Challenge was to super-resolve combined diffusion-relaxometry MRI data.
The challenge consisted of two tasks, with each task exploring a different low-resolution MRI acquisition strategy, leading to two different types of images to super-resolve.
The paper ‘Deep Learning-based Fetoscopic Mosaicking for Field-of-View Expansion’, including authors from the School Professor Sebastien Ourselin and Professor Tom Vercauteren was awarded the IJCARS MICCAI 2019 Special Issue Best Paper Award.
A team from the Robotics and Vision in Medicine (RViM) lab, a research group within the School, consisting of PhD students Theodoros Pissas, Claudio Ravasio, Martin Huber, Jeremy Birch, and Postdoc Joan Nunez do Rio, won the MICCAI 2020 CATARACTS Semantic Segmentation Challenge – first out of 11 teams.
The competition required participants to label each pixel in a number of frames taken from cataract surgery videos as belonging to a specific anatomy, or type of tool, in a series of three increasingly difficult tasks, with the ultimate goal of empowering computer assisted interventions by aiding detailed post-operative analysis.
“Our approach as a team was to develop a code base that enabled quick parallel experimentation with a large number of state-of-the art methods for deep learning-based semantic segmentation,” the researchers said.
They focused on experimentally selecting an effective convolutional network architecture and addressing the discrepancy between the loss functions typically used to train the models and the evaluation metric used to assess performance.
They placed significant effort in addressing the dataset class imbalance that is a very common challenge in real-world computer vision applications where unavoidably some objects like some surgical tools, appear much less frequently.
“Our collaborative approach allowed us to later combine different elements that proved effective in a thoroughly optimized model that gave us the 1st place in the two most challenging out of the three sub-tasks, while achieving 3rd place in the simpler sub-task,” they said.
PhD student Lucas Fidon, who researches robust deep learning methods for the segmentation of medical images, made it in the top performing methods at the Multimodal Brain Tumor Segmentation challenge with a WassersteinDice entry.
Another PhD student, Helena Williams from KU Leuven and affiliated to King’s College London, who researches automation of pelvic floor disorder assessment by utilising advances in deep learning made second place for the best presentation award at Advances in Simplifying Medical UltraSound (ASMUS) workshop.
“This study is clinically driven to aid clinicians within pelvic floor disorder assessments and is in collaboration with GE Healthcare. It was a great workshop to attend, the focus on clinical need and clinical impact was very strong and I very much look forward to next year,” she said.
The annual MICCAI conference attracts world leading biomedical scientists, engineers, and clinicians from a wide range of disciplines associated with medical imaging and computer assisted intervention.