by John R. Fischer
, Senior Reporter | April 25, 2019
Researchers at the University of California, Los Angeles have developed a new AI system with a level of efficiency comparable to that of longtime radiologists in detecting prostate cancer.
Deemed FocalNet, the solution can identify and predict the aggressiveness of the disease, based on MR scans, with nearly the same amount of accuracy as radiologists with 10 years of experience. Its developers argue that the system could aid hospitals that are limited in their ability to supply the correct training necessary for radiologists to learn how to accurately identify benign and cancerous tumors and the grade of malignancies with multi-parametric MR.
“Multi-parametric MR includes T2-weighted (T2w) MR, diffusion-weighted imaging (DWI) and dynamic contrast-enhanced MR (DCE-MR),” Dr. Kyung Sung, assistant professor of radiology at the David Geffen School of Medicine at UCLA, told HCB News. “This requires many years of experiences and volume to train, which may not be easy to implement for many hospitals.”
Numed, a well established company in business since 1975 provides a wide range of service options including time & material service, PM only contracts, full service contracts, labor only contracts & system relocation. Call 800 96 Numed for more info.
An artificial intelligence network, FocalNet utilizes an algorithm made up of more than one million trainable variables.
The researchers trained the system to assess and classify tumors consistently by feeding it MR scans from 417 prostate cancer patients and compare its findings to the actual pathology specimen.
Comparing its results with those of UCLA radiologists with more than 10 years of experience, the team found FocalNet to be accurate 80.5 percent of the time, compared to the radiologists at 83.9 percent.
They believe that with more training, the system could save on time and potentially become a diagnostic guide for less experienced radiologists.
“We anticipate the system would be improved when we include more training data sets, particularly from different MR scanners, and integration with patients' clinical information, such as clinical history and PSA scores,” said Sung. “It also can [detect] other cancers with MR, and we plan to expand its use to breast multi-parametric MR.”
The findings were published in IEEE Transactions on Medical Imaging
, and were presented this month at the the IEEE International Symposium on Biomedical Imaging.