Strong UCL presence at MICCAI 2017
The 20th international conference of Medical Image Computing and Computer Assisted Intervention (MICCAI) took place in Quebec, Canada on 10 September – 14 September 2017.
There was a strong UCL presence at the conference with a number of oral presentations, poster presentations and contributions at associated satellite events from our researchers. The Workshops Co-Chair, part of the main conference organising committee, was Dr Jorge M. Cardoso who works within the Institute of Healthcare Engineering and Translational Imaging Group (TIG).
Guotai Wang, a PhD student at the EPSRC Centre for Doctoral Training in Medical Imaging, came 2nd place out of 60 teams in the MICCAI Multimodal Brain Tumour Segmentation (BraTS) challenge for his method of Automatic Brain Tumour Segmentation using Cascaded Anisotropic Convolutional Neural Networks. The developed a deep-learning network using NiftyNet which is able to automatically detect and segment brain tumours from standard MRI imaging data. The impact of such a process would greatly aid better diagnosis, surgical planning and treatment assessment for brain tumours.
Other successes that stemmed out of work from within the Institute included Catherine Scott’s poster presentation of her research into the early detection of Alzheimer’s Disease (AD) using simultaneous PET/MRI scans. This advanced technique allows researchers to measure the number of sticky protein lumps deposited in the brain, which can be a good indicator of future brain atrophy and AD.
Lucas Fidon presented research aiming to tackle a problem that has limited the application of software that automatically and accurately segments brain tumours: that models designed to work with one imaging modality tend to generalise poorly to different imaging modalities. To combat this, he proposes a new, deep learning model that has sparse and structured early layers, therefore making the framework scalable for use with other modalities.
Other presentations touched on a wide variety of subjects within medical imaging. Juan Eugenio Iglesias’s oral presentation focused on enhancing the accurate estimation of intracranial volume in preterm birth animals. Notably, his method uses an algorithm, meaning images do not need to be manually labelled by a human expert which can be time-consuming.
Further, David Owen presented a method to better process the data measuring blood flow in the brain. By reducing some of the interference levels created through Arterial Spin Labelling (ASL) he was able to produce higher-quality images with a better contrast. In the validation data that they used in the paper, they found that the method could detect AD significantly more reliable than pre-existing approaches, therefore offering great promise for future neuroimaging studies.
One of the main aims of MICCAI 2017 was to foster the exchange and dissemination of advanced knowledge, expertise and experience in the field, and this year’s edition attracted over 1,200 attendees. As such, the conference provided important networking opportunities.
This was frequently commented on by the UCL researchers who attended MICCAI, for instance David Owen said, “I met a number of people working on similar blood flow imaging problems, and we were able to discuss sharing some of our software tools”. It is hoped that the connections made will continue to grow and that this collaborative spirit will continue at MICCAI 2018 in Granada, Spain.