Can a statistical model find connections between imaging and genomic measures in brain cancer?


The time-intensive and costly process of getting a biopsy isn’t ideal for treating fatal brain cancers. Low-grade glioma, a type of slow-growing brain tumor, requires early detection and treatment to extend a person’s life. Researchers from Boston University, University of Michigan, Ohio State University, and University of Nottingham developed a tool that could potentially speed up the detection and enable treatment planning of low-grade gliomas.

“If we can identify associations between images and specific genomic markers which are known to relate to long- or short-term survival, we can enable early treatment planning and management,” says Shariq Mohammed, a Junior Faculty Fellow at the Hariri Institute and Assistant Professor in Biostatistics at the School of Public Health.

Mohammed presented his findings recently at the Royal Statistical Society (RSS) International Conference. Using validated genetic markers and radiological information, the team’s statistical model could be a non-invasive complementary approach for diagnosing and monitoring low-grade gliomas.

The researchers trained their model using three-dimensional magnetic resonance imaging (MRI) scans from The Cancer Imaging Archive and matching genomic information obtained from biopsies. First, the team divided the imaged 3D tumors into three spherical layers to mimic the tumor evolution process. Then, the researchers trained their model to associate characteristics of the tumor’s texture from the three layers, with genomic indicators of cancer from The Cancer Genome Atlas. The researchers then tested whether their model could identify cancerous genes as being associated with MRI scans of low-grade gliomas.

The researchers discovered that their model was able to associate known markers of cancer with images of tumors. Genes associated with the inner layers of the tumor, where cells start to die, could serve as early indicators of cancer for speedier diagnoses. The team’s findings suggest that the physical features of tumors provide effective diagnostic options to monitor cancer before biopsies are obtained. The researchers were also surprised that their model might provide insight into other genes that might not be known links, yet, to brain cancer. “There are some markers that have not been highlighted in the literature for brain cancer that our method is highlighting as potential biomarkers. A deeper validation of these is essential to better understand tumor etiology,” says Mohammed.

The team’s statistical model uses methods that can be generalized and applied to other cancers, diseases, or conditions. Mohammed’s current research at BU leverages the tool in a new context: he uses data from digital pens to determine whether individuals are at risk of developing Alzheimer’s Disease and dementia.  “I am a trained statistician, and I am working in a space to develop models that make meaningful inferences and can impact a patient’s health,” says Mohammed.

To learn more about the Hariri Institute’s transformational research, click here to sign up for our newsletter.