Intelligent imaging informatics for surgery

点击翻译按钮获取中文摘要

Brain surgery to treat drug-resistant epilepsy or brain cancer is very challenging, as the surgeon needs to trade-off completely removing the disease, with sparing healthy functional tissue. Planning how much of the brain to treat requires knowledge of the neurological pathways and abnormal regions, however this can be difficult as the boundaries are not always well defined. Incorrectly defining the region to treat can carry a great deal of risk; if too small, the surgery will not be effective, and if too large, function may be altered negatively. Magnetic resonance imaging (MRI) has attempted to solve these problems non-invasively, however, the puzzle piece that has been missing is the integration of signal information that is collected during and after surgery. Clinicians commonly rely on this signal information, such as functional recordings from electrodes placed on the brain, or microscopic images of tissue, to confirm what can only be suspected with MRI. Fusing these biomedical signals with MR imaging closes the loop, and allows us to gain a deeper understanding of how image analysis can be used to precisely highlight the target. Gaining an understanding of this fused biomedical information requires us to address three key challenges: 1) robustly extracting information on what neurological pathways in the brain are functional and need to be avoided, 2) finding spatial correspondence between images taken before, during, and after surgery, and 3) using all this information, built into an informatics database, to outline the target in a new patient and the uncertainty in this prediction. The goal of the proposed research program is to develop sophisticated software tools that address these challenges. This work will be the first to systematically align and fuse functional and biological information from surgery with pre-operative images, and will allow for an unprecedented ability to compare this information at a local scale. The contributions in computational image analysis will also be of significance to the wider biomedical image processing community, as they address many key challenges related to robustness in image registration and how uncertainty is modelled. Ultimately, this work will transform how image-guided surgery is performed, and will pave the way for the delivery of more precise and minimally-invasive treatment.