UCLA Radiology researchers, led by Ruiming Cao, Kyung Sung, PhD and Steven Raman, MD, have developed a new artificial intelligence application — FocalNet — that detects prostate cancer lesions and predicts their aggressiveness on multi-parametric magnetic resonance imaging (mp-MRI) scans. With comparisons to the prospective performance of radiologists using the current diagnostic guideline, FocalNet demonstrated nearly the same level of accuracy as radiologists with 10 years of clinical prostate MRI reading experience.
FocalNet is a novel multi-class convolutional neural network (CNN) that automatically learns to assess and classify prostate lesions in a consistent way. The system learned by analyzing mp-MRI scans of 417 men with prostate cancer and comparing its own results to the actual pathology specimen.
The researchers compared FocalNet with the prospective clinical performance of three highly-experienced genitourinary radiologists for lesion detection and did not find statistically significant differences. As prostate MRI reading quality largely varies according to the radiologist’s experience, FocalNet can potentially assist both experienced and less experienced radiologists or augment the prostate cancer detection task for non-experts to minimize diagnostic errors on MRI.