November-December 2017
Expedited Publication
Figure from Perreault

MRI Surrogates for Molecular Subgroups of Medulloblastoma

Yeom Pic

Kristen Yeom

Back in February 2008, Scott Atlas, Paul Fisher, and I sat together brainstorming how we might tackle tricky cases of patients with pediatric medulloblastomas. This was a time before molecular subtyping of medulloblastoma. We all knew medulloblastomas could look like anything. They could sit in intra- and/or extra-axial spaces. They could look like a ball, a beehive, a blob from outer space, or simply a cyst and a mural nodule—just like the pilocytic tumors we learned about back in residency before diffusion came to the rescue. They could also behave differently—at least that is what Paul, our luminary pediatric neuro-oncologist, told us. Some children succumbed to their disease quickly, whereas others were already applying for college.

So we agreed that something must have been up at the molecular level and that it would be great if we could sort these tumors out molecularly and put them together with image features predictive of tumor behavior or clinical outcome (ie, radiogenomics).  Luckily for us, sometime around 2010, the Michael Taylor lab in Toronto and the Scott Pomeroy lab in Boston independently identified clinically significant medulloblastoma molecular subtypes. We were excited to partner with these groups and further investigate image features that might correlate with underlying molecular subtypes of pediatric medulloblastoma. In this work led by Dr. Sebastien Perreault, we found that there were indeed important qualitative, or semantic, image features that reflected underlying molecular events. As a result of recognizing that a robust predictive model based on subtype-specific MRI features could further assist in clinical risk stratification, we are currently working with teams in Toronto, Boston, and Stanford on a computational radiogenomics strategy for approaching pediatric medulloblastoma. We have thus far extracted 500 computational image features and identified specific radiomics features predictive of medulloblastoma subtypes. It is an exciting time for radiogenomics research, and we hope to share our exciting results in the coming months.

Read this article at …