After training on nearly 15,000 3D OCT scans, a deep-learning algorithm developed by DeepMind was able to segment and then analyze these studies for over 50 retinal diseases. In testing, the algorithm yielded referral suggestions that were either equivalent or exceeded those made by retinal specialists and optometrists participating in the study.
"Our framework can triage scans at first presentation of a patient into a small number of pathways used in routine clinical practice with a performance matching or exceeding both the expert retina specialists and optometrists who staff virtual clinics in a U.K. National Heath Service setting," wrote the researchers, led by Jeffrey De Fauw.
The algorithm achieved an area under the receiver operating characteristic (ROC) curve of 0.99 for most pathologies, and over 0.96 for all pathologies. Those results were on par with the performance of experts viewing only the OCT studies; their performance improved, however, when they were also provided fundus image and patient summary results, the researchers noted.
The algorithm's error rate for referral decisions was 5.5%, comparable with expert performance. In testing of images generated on a different type of OCT device, the model produced an error rate of 3.4%. Again, the difference in the algorithm's performance from the retina specialists was not statistically significant, according to the researchers.
"Although we focused on one common type of medical imaging, future work can address a much wider range of medical imaging techniques, and incorporate clinical diagnoses and tissue types well outside the immediate application that was demonstrated here," the authors concluded.
Copyright © 2018 AuntMinnieEurope.com