High Stakes Hide-and-Seek
Can artificial intelligence help surgeons and radiation oncologists find invisible brain tumors? Scientists at MCW and their peers have discovered the presence of cancerous tissue in some research volunteers that is undetectable by even the most sophisticated medical imaging techniques.
To better understand this problem, Drs. Peter LaViolette and Sam Bobholz worked with a team of collaborators in multiple MCW departments and two California medical schools to obtain and study 159 tissue samples from 65 deceased patients who had suffered from brain cancer and volunteered to participate to advance detection and treatment for future generations.
Drs. LaViolette, Bobholz and team then used these patients’ MRI scans to determine precise measurements and print unique 3-D molds to match each brain. This mold allows the brain to be sliced and processed into glass slides that accurately match the corresponding MRI scan. The team’s pathologists analyzed the tissue for tumor severity and location before sending annotated images to Drs. LaViolette and Bobholz. Drs. LaViolette and Bobholz then employed a machine learning program to read and compare the hundreds of pathologist-annotated slides and the MRI data.
Machine learning is a rapidly growing and evolving approach to analyzing and interpreting massive amounts of data. Unlike in traditional programming that requires scientists to provide a computer with every parameter of a problem using handwritten code, machine learning developers and scientists apply artificial intelligence to set up frameworks through which computers learn on their own. By comparing the many images provided by the scientists, the software learns what features in the MRI are predictive of the tumors found in the annotated slides.
Drs. LaViolette, Bobholz and team reported in a preprint of the upcoming manuscript that they have successfully identified previously invisible tumor in 72.5 percent of research subjects with their tumor probability maps. This demonstrates that a significant subset of brain cancer patients may one day benefit from clinical application of this mapping technique once further research has been conducted. By revealing these otherwise hidden boundaries of tumors, cancer progression may be able to be monitored with greater precision and future surgeries and targeted radiation treatments may become more effective, reducing the chance of recurrence.
Contributors to this Project
Research in Collaboration With:
Allison Lowman, BS, MCW Department of Radiology
Jennifer Connelly, MD, MCW Department of Neurology
Fitzgerald Kyereme, BS, MCW Department of Radiology
Savannah Duenweg, BS, MCW Department of Biophysics
Aleksandra Winiarz, BS, MCW Department of Biophysics
Michael Brehler, PhD, MCW Department of Radiology
Elizabeth Cochran, MD, MCW Department of Pathology
Janine Lupo, PhD, University of California-San Francisco Department of Radiology and Biomedical Imaging
Joanna Phillips, MD, PhD, University of California-San Francisco Departments of Neurological Surgery and Pathology
Benjamin Ellingson, PhD, University of California-Los Angeles David Geffen School of Medicine Department of Radiological Sciences, Brain Tumor Imaging Laboratory
Wade Mueller, MD, MCW Department of Neurosurgery
Mohit Agarwal, MD
Anjishnu Banerjee, PhD, MCW Institute for Health & Equity Division of Biostatistics