AI imaging technique accurately analyzes genetic mutations in brain tumors

Potential for noninvasive virtual biopsy may lead to faster, more personalized tumor treatment

July 30, 2018

Researchers at the University of California, Irvine have developed an artificial intelligence-based imaging technique that accurately analyzes genetic mutations in brain tumors and makes possible the use of virtual biopsies. The findings advance a type of machine learning called convolutional neural networks and demonstrates the capability to recognize key imaging details without human direction.

The paper, “Deep-Learning Convolutional Neural Networks Accurately Classify Genetic Mutations in Gliomas,” is available online and in the July 2018 American Journal of Neuroradiology.

“Our aim was to train a convolutional neural network to independently predict molecular genetic mutations in gliomas with high accuracy and classify the features that are most associated with each kind of mutation,” said Daniel S. Chow, MD, assistant professor of radiological sciences, UCI School of Medicine. “The findings suggest that noninvasive, image-based ‘virtual biopsies’ are possible and may be a valuable complement to conventional tissue sampling in diagnosing and identifying personalized treatment options for some brain tumors.”

Chow is co-director, with neuroradiologist Peter D. Chang, MD, of the UCI Center for Artificial Intelligence in Diagnostic Medicine.

“We achieved 94 percent diagnostic accuracy in determining relevant genetic mutations in MR images from 256 patients with either low- or high-grade gliomas” said Chang, assistant professor of radiological sciences.

The imaging data and molecular information came from prospective studies in National Cancer Institute's The Cancer Imaging Archive and The Cancer Genome Atlas, respectively.
Molecular analysis of tumors through tissue biopsy has significantly improved the diagnosis and treatment of gliomas. However, many high-grade gliomas, such as the deadly glioblastoma, infiltrate deeply into the brain, often making surgical biopsy difficult or incomplete.

Magnetic resonance imaging offers a noninvasive view of the entire tumor, allowing a multidimensional assessment of its molecular genetics and potential mutations that may have a significant bearing on how the patient’s disease is treated. The conventional assessment of tissue biopsy is spatially limited and can sometimes add delays to treatment, as molecular genetic testing may be costly or not widely available, and the results may take weeks.

Deep learning techniques have a reputation for being ‘black boxes,’ making it difficult to assess how the technology determines its classification. However, the UCI group has developed approaches that visualize aspects of the network and identify clusters of features that are clinically relevant, a step towards biomarker discovery. This approach improves confidence in the network while also allowing for identification of new biomarkers and imaging phenotypes. 

Moreover, these findings demonstrate that neural networks are capable of learning key imaging components without prior feature selection or human-directed training.

This research was supported, in part, by the following grants: Defense Advanced Research Projects Agency D17AP00002 and National Institutes of Health GM123558 and NIH T32EB001631. Canon Medical Systems USA provided additional support.

UCI Health comprises the clinical enterprise of the University of California, Irvine. Patients can access UCI Health at primary and specialty care offices across Orange County and at its main campus, UCI Medical Center in Orange, California. The 417-bed acute care hospital provides tertiary and quaternary care, ambulatory and specialty medical clinics, and behavioral health and rehabilitation services. UCI Medical Center features Orange County’s only National Cancer Institute-designated comprehensive cancer center, high-risk perinatal/neonatal program and American College of Surgeons-verified Level I adult and Level II pediatric trauma center and regional burn center. It is the primary teaching hospital for the UCI School of Medicine. UCI Health serves a region of nearly 4 million people in Orange County, western Riverside County and southeast Los Angeles County. Follow us on Facebook and Twitter.

About the University of California, Irvine: Founded in 1965, UCI is the youngest member of the prestigious Association of American Universities. The campus has produced three Nobel laureates and is known for its academic achievement, premier research, innovation and anteater mascot. Led by Chancellor Howard Gillman, UCI has more than 30,000 students and offers 192 degree programs. It’s located in one of the world’s safest and most economically vibrant communities and is Orange County’s second-largest employer, contributing $5 billion annually to the local economy. For more on UCI, visit www.uci.edu.