MICCAI 2019 Tuesday

MICCAI 2019 DAILY 3 Oral Presentation - Bo Li Bo Li is a postdoc researcher in the BIGR group at the Erasmus Medical Center in theNetherlands. Shespeaks to us following her poster session yesterday and ahead of today’s oral, where she is presenting work on a hybrid framework integrating both segmentation and registration in a single convolutional network. Her supervisor for this work is Wiro Niessen, former president of the MICCAI Society. Bo tells us that segmentation and registration are two tasks that are essential in many applications, both separately and combined. In a longitudinal study where you want to analyze brain changes over time, and you need to quantify brain changes and do longitudinal alignments, these two tasks can now be integrated into a single procedure with improved performance. Not only improving the segmentation accuracy and registration accuracy, but most importantly, optimizing the segmentation consistency that relies on both. This work has been applied to quantify white matter tract changes over time based on the large-scale Rotterdam Studydataset,whichhas8,000diffusion- weightedMRI scans. Bo says they would like to generalize this joint optimization to other applications. For example, to integrate anatomical segmentation with disease classification and make it a single procedure, or to integrate motion correction with image reconstruction. There are already some people in her group doing that. Bo tells us more about future plans: “We would like to evaluate this method more in clinical applications. First, we would like to evaluate its applicability in other modalities.Now, it’s indiffusion-weighted MRI, but as a next step we would like to evaluate structural MRI to segment gray matter changes. Also, to put it in a real pre-clinical and clinical evaluation to analyze the brain changes.” To learn more about this work, come along to Bo Li’s oral today at 14:16. A hybrid deep learning framework for integrated segmentation and registration: evaluation on longitudinal white matter tract changes

RkJQdWJsaXNoZXIy NTc3NzU=