Researchers at Rutgers, the State University of New Jersey, report that they have achieved greater than 90% accuracy in distinguishing high-grade from low-grade prostate cancers using computer analysis of magnetic resonance (MR) images and spectra of the patient’s prostate gland. The findings may help determine which patients need aggressive treatment and which may be better served by watchful waiting.
Earlier research used high-resolution MR imaging technology to detect prostate cancer. ”Now we’re getting beyond merely identifying whether a person has cancer or not,” said Anant Madabhushi, associate professor of biomedical engineering at Rutgers and a member of The Cancer Institute of New Jersey.
"The breakthrough we’ve had in the last few months is that we see image signatures that distinguish aggressive cancers from less aggressive ones,” he explained. “This could lead to better patient management and cost savings.”
The researchers obtained prostate gland images from 19 men who later had radical prostatectomies. They examined both traditional MR images, which provide two-dimensional pictures of the gland’s cellular structure and MR spectroscopy, which maps concentrations of certain chemicals (choline, creatine, and citrate) to locations in the prostate gland. Changes in levels of these chemicals indicate the presence of cancer.
The researchers compared the MR images and spectra with digital images of the actual excised prostate glands, which pathologists had identified as having high-grade or low-grade tumors according to the Gleason Grading System. Using pattern recognition techniques, they identified characteristics of areas in the MR images and spectra that corresponded to cancerous tissue in the excised samples. The objective, the researchers said, is to “teach” the computer to accurately and consistently recognize image patterns that correspond to various grades of cancerous tissue without having the tissue samples available for examination.
Madabhushi called the early findings encouraging but said that further study is needed before the image analysis techniques can be considered for clinical application.
The researchers will present their findings at the Medical Image Computing and Computer Assisted Intervention Conference in Beijing, China, in September.