Using Brain Tumor Geometry & Topology to Predict Clinical Outcomes

While brain cancer is one of the rarest types of cancer, it can have a devastating impact on an individual's behavior and every-day functioning.  Gliomas are a type of brain tumor that are found to exist in one of several states.  Low-grade gliomas (LGG) grow very slowly and often do not require surgery; however, if they transition into a medium- or high-grade glioma they must be removed as soon as possible.  Thus a sensitive and easy to use method is needed to track the progress of LGG tumors and identify if and when they require resection.  The gold-standard measure of glioma type is laboratory analysis of the tumor tissue, which requires surgery; thus there is great interest in using non-invasive imaging, like Magnetic Resonance Imaging (MRI) to predict the tumor status.

The standard measure of a brain tumor size is just the largest diameter as measured one radiographic image. This low-tech approach doesn't capture the total tumor size, and doesn't reflect whether it has a smooth or irregular shape, and doesn't use the multiple types MRI scans collected.  The goal of this project is to obtain geometrical and topological summaries of the shape of the tumor, as well as information on the intensities of the MR image inside the tumor, to predict the grade of the tumor.

The project will use supervised learning methods to predict true tumor status. The image summary measures and geometry/topology measures are already computed, but more measures can be considered by the student depending on the interest and capability with image data.  The end result of this project will contribute to more accurate clinical care of LGG patients.

Superviors:  Tom Nichols, Andreas Bartsch (Heidelberg University).