Apr 17 2019
A research team at UCLA has created a novel artificial intelligence system that can aid radiologists to enhance their potential to diagnose the aggressive disease, prostate cancer.
Authors Ruiming Cao, Dr Steven Raman and Kyung Sun. (Image credit: UCLA)
Known as FocalNet, the system helps in identifying and predicting the aggressiveness of prostate cancer by assessing MRI (magnetic resonance imaging) scans. The system reads MRIs with almost the same level of precision as experienced radiologists. Tests were conducted in which FocalNet was shown to be 80.5% accurate when it came to reading MRIs, whereas radiologists with a minimum of one decade of experience were only 83.9% accurate.
Background
MRI is used by radiologists for detecting and evaluating the aggressiveness of malignant prostate tumors. Yet, most often, radiologists have to practice on a countless number of scans to learn the best way to precisely establish whether a tumor is indeed benign or cancerous, and then estimate the grade of cancer in an accurate manner. Moreover, a majority of hospitals lack the resources to apply the highly specialized training needed for identifying cancer from MRIs.
Method
The artificial neural network—FocalNet—employs an algorithm that contains over a million trainable variables; it was created by the UCLA team. To train the system, the researchers allowed it to examine MRI scans of 417 prostate cancer patients; scans were subsequently fed into the system as this would allow it to evaluate and categorize tumors in a reliable manner and it can then compare the outcomes to the real pathology specimen. The team then compared the results of the artificial intelligence system with the readings taken by UCLA radiologists who had over a decade of experience.
Impact
According to the study, an artificial intelligence system may save a considerable amount of time and probably offer diagnostic guidance to radiologists who are less experienced.
Authors
Senior authors of the study are Kyung Sung, assistant professor of radiology at the David Geffen School of Medicine at UCLA; Dr Dieter Enzmann, chair of radiology at UCLA; and Dr Steven Raman, a UCLA clinical professor of radiology and a member of the UCLA Jonsson Comprehensive Cancer Center. UCLA graduate student Ruiming Cao is the lead author of the study. Other authors include Sohrab Mirak, Amirhossein Bajgiran, Xinran Zhong, and Sepideh Shakeri, all from UCLA.
Journal
The study has been reported in IEEE Transactions on Medical Imaging. In April 2019, the paper was presented at the IEEE International Symposium on Biomedical Imaging and was chosen as the runner up for best paper.