Road and parking lot construction in Madison, Wis. may result in travel delays and route changes to UW Health clinic and hospital locations. Please plan accordingly.Read more
Glioblastoma is an aggressive form of brain cancer that has a high recurrence rate even after surgery and radiation treatments.
Right now, imaging scans cannot tell the difference between what may be new tumor growth, and what may be a noncancerous lesion caused by radiation. As a result, patients need to go through another surgery so that either the tissue can be sampled for a biopsy, or the questionable area is removed completely.
“Ultimately, in 40 percent of cases (surgeons) find out that it was a benign lesion and should not have been taken out,” said Dr. Pallavi Tiwari, visiting associate professor in the University of Wisconsin Departments of Radiology and Biomedical Engineering, and co-director of Imaging and Radiation Science at UW Carbone Cancer Center.
Tiwari, who came to UW last summer, aims to minimize the need for these exploratory surgeries through use of artificial intelligence models that can effectively diagnose glioblastoma recurrence or benign lesions.
Tiwari’s expertise lies in developing machine learning with medical imaging to improve patient diagnostics and tracking of disease progression and treatment effectiveness. Her work has received significant grant support, including nearly $4 million from the National Institutes of Health to develop her models for glioblastoma recurrence. She also recently was selected as part of the 2023 Class of Senior Members of the National Academy of Inventors.
Tiwari has spent eight years developing artificial intelligence models that examine hundreds of brain scans to identify patterns and nuances that can improve treatment outcomes in adult and pediatric brain tumor patients. These images are sourced from existing patient scans at medical centers nationwide.
“With building these models, you need large datasets, multi-institutional datasets to ultimately have a model that’s going to be accurate, robust, and has a wider footprint,” Tiwari said. “And thus, our models are developed very carefully to not only be accurate but also be able to account for the differences in imaging scans from across different sites/scanners.”
Before these modeling programs can be approved for clinical use, they must be thoroughly vetted for accuracy compared to real patient cases. Tiwari now is working with GE Healthcare to pilot these models with neuroradiologists at three medical centers in the U.S, as a part of her ongoing NIH grant.
Tiwari’s lab also focuses on using machine learning to evaluate scans of disease progression for patients with adult and pediatric brain cancer, specifically examining which treatments were effective for specific tumor types, and which were not in order to develop models that can help place patients in the clinical trials most likely to benefit them.
“It’s challenging just because every person’s tumor is different, and the way they respond to treatments is different,” she said.
While Tiwari’s main focus has been on adult and pediatric brain cancers, she said this approach can be expanded and adapted for study across several types of cancer.
“All of the techniques that we have been working on, that we’ve been developing, are fairly generalizable, and can be extended to developing personalized medicine solutions for other solid tumors as well,” she said.