Two recent grants totaling $492.000 from the National Institutes of Health, a $50.000 grant from the New Jersey Commission on Cancer Research and a $25.000 grant from the Charles and Johanna Busch Biomedical Foundation build on initial grants of $255.000 received last year from the Coulter Foundation.
"Each year, there are more than 40.000 deaths from prostate cancer in the United States and 240.000 new cases diagnosed", stated principal investigator Anant Madabhushi, assistant professor of biomedical engineering at Rutgers and member of the Cancer Institute of New Jersey.
Physicians who suspect prostate cancer based on high levels of prostate specific antigen (PSA) in the blood typically order a needle biopsy, because common imaging techniques, such as ultrasound, don't distinguish cancerous tissue. Biopsies are difficult to perform because ultrasound can't reveal the precise location of suspicious tissue. While physicians take several samples from the prostate, a localized mass of cancerous tissue could still be missed.
"This results in biopsy accuracy rates of only 20 to 25 percent even for patients with high levels of PSA", stated Anant Madabhushi. The researchers are using a three tesla MRI machine at the University of Pennsylvania, which is twice as strong as machines typically found in clinical settings, to reveal masses of suspicious tissue. Once found, they apply computer-aided diagnostics to predict the likelihood that it is cancerous based on an image's size, shape and texture.
"In our testing, MRI images revealed several spots of suspicious tissue, but four expert radiologists disagreed as to which ones were cancerous", stated Anant Madabhushi. "Our computer-aided techniques assign probability ratings to the spots, which help a radiologist reach a more confident answer."
A physician would still need to confirm a radiologist's diagnosis with a biopsy. Still, the techniques being developed could eliminate biopsies when suspicious tissue is judged to be healthy. The new methods also could help physicians home in on the tissue in question, resulting in more accurate biopsies.
The same imaging techniques could prove useful in treatment. A definitive fix on where cancerous tissue lies could help physicians focus radiation on that region and reduce collateral damage to neighbouring tissues and organs.
The researchers are also applying CAD techniques to histology, or the detailed analysis of tissue samples taken during biopsies. Just as radiologists differ in their interpretation of prostate images, pathologists may assign different severity grades to the same sample of cancerous tissue. Inaccurate grading may result in therapy too weak to treat the cancer or so aggressive that it prolongs therapy and harms healthy tissue. Early results show that where pathologists have trouble distinguishing between two medium grades of severity, computer-aided analyses can deliver an 80 percent accuracy rate.
Another technique the Rutgers and Penn scientists are pursuing is magnetic resonance spectroscopy (MRS), which measures the concentrations of chemicals in the prostate gland. Healthy tissue shows different proportions of chemicals than cancerous tissues, so MRS might provide more convincing evidence for the presence of disease. Yet, as with imaging, human interpretations of the findings can be ambiguous, so the researchers are applying CAD to increase the accuracy of interpretation here, as well.
In a paper to be presented at the prestigious International Conference on Medical Image Computing and Computer Assisted Intervention this autumn, Anant Madabhushi and his collaborators will describe how sophisticated machine learning tools identify potential cancerous locations within the prostate based on MRS signatures.
Members of the research team from the University of Pennsylvania School of Medicine are Dr. Michael Feldman, assistant professor of pathology and laboratory medicine; Dr. Mark Rosen, assistant professor of radiology; and Dr. John Tomaszeweski, professor of surgical pathology. Rutgers graduate student researchers include Jonathan Chappelow, Scott Doyle, Pallavi Tiwari, and Satish Vishwanath.