As the influence of artificial intelligence continues to grow, researchers are finding more and more new ways to take advantage of convolutional neural networks (CNNs) in healthcare. According to a new study published in Radiology, using a CNN as a deep learning algorithm can help improve the overall quality of arterial spin labeling (ASL) image quality.
“ASL perfusion images usually are acquired by averaging a number of pair-wise subtractions, requiring a longer imaging time,” wrote lead author Ki Hwan Kim, MD, Korea Advanced Institute of Science and Technology in Daejeon, South Korea, and colleagues. “To reduce imaging time, we aimed to develop CNNs that generate ASL perfusion images with higher accuracy and robustness by using a smaller number of pair-wise subtractions.”
The authors compared the CNN’s performance with the traditional averaging method used in ASL imaging. Overall, the CNN “provided perfusion images much closer to those of the ground truth,” the authors found. In addition, using the CNN resulted in fewer mean square errors and higher radiologic scores.
The CNN also reduced segmentation and motion artifacts visible when using conventional ASL imaging methods.
“In conclusion, CNN showed superior performance with ASL perfusion imaging compared with the conventional averaging method, regardless of MR imager, labeling scheme, and readout scheme,” the authors wrote. “The performance of CNN was demonstrated for data sets from a separate cohort of patients who had experienced a stroke.”