Identification of Clinically Occult Glioma Cells and Characterization of Glioma Behavior Through Machine Learning Analysis of Advanced Imaging Technology

Overview

Gliomas are one of the most challenging tumors to treat, because areas of the apparently normal brain contain microscopic deposits of glioma cells; indeed, these occult cells are known to infiltrate several centimeters beyond the clinically apparent lesion visualized on standard computer tomography or magnetic resonance imaging (MR). Since it is not feasible to remove or radiate large volumes of the brain, it is important to target only the visible tumor and the infiltrated regions of the brain. However, due to the limited ability to detect occult glioma cells, clinicians currently add a uniform margin of 2 cm or more beyond the visible abnormality, and irradiate that volume. Evidence, however, suggests that glioma growth is not uniform - growth is favored in certain directions and impeded in others. This means it is important to determine, for each patient, which areas are at high risk of harboring occult cells. We propose to address this task by learning how gliomas grown, by applying Machine Learning algorithms to a database of images (obtained using various advanced imaging technologies: MRI, MRS, DTI, and MET-PET) from previous glioma patients. Advances will directly translate to improvements for patients.

Gliomas are the most common primary brain tumors in adults; most are high-grade and have a high level of mortality. The standard treatment is to kill or remove the cancer cells. Of course, this can only work if the surgeon or radiologist can find these cells. Unfortunately, there are inevitably so-called "occult" cancer cells, which are not found even by today's sophisticated imaging techniques. This proposal proposes a technology to predict the locations of these occult cells, by learning the growth patterns exhibited by gliomas in previous patients. We will also develop software tools that help both practitioners and researchers find gliomas similar to a current one, and that can autonomously find the tumor region within a brain image, which can save radiologists time, and perhaps help during surgery.