Artificial intelligence can be as effective as radiologists—or more so—in detecting acute intercranial hemorrhages (ICH) on head CT scans, according to a study published online October 21 in the Proceedings of the National Academy of Sciences.
Researchers from the University of California, San Francisco (UCSF) and the University of California, Berkeley (UC Berkeley) drew this conclusion by developing and testing a single-stage, fully convolutional, end-to-end neural network model called PatchFCN. The deep learning model was trained on 4,396 CT studies in which individual small abnormalities were manually delineated at the pixel level.
The researchers opted against training the algorithm on either a single image or stack of images. Rather, they utilized patches of images that had been contextualized with those that directly preceded and followed them in the stack of images being examined. This approach, they wrote, enabled PatchFCN to learn from the relevant information in the imaging data without reaching conclusions based on insignificant variations within that data.
PatchFCN’s performance was compared with the performance of four expert radiologists on a separate test set of 25 positive and 175 negative head CT exams. In addition to an area under the curve (AUC) of 0.991 ± 0.006, the algorithm identified some small abnormalities not pinpointed by the radiologists.
"Algorithm performance exceeded that of 2 of 4 American Board of Radiology (ABR)-certified radiologists with attending-level experience ranging from 4 to 16 (years)," wrote study author Esther Yuh, PhD, of UCSF and colleagues. "In addition, PatchFCN achieved 100% sensitivity at specificity levels approaching 90%, making this a suitable screening tool for radiologists based on an acceptably low proportion of false positives."
Yuh et al also noted that the five cases deemed negative by at least two of the four radiologists, but positive by the algorithm and the gold standard alike, contained very tiny ICHs that are more likely to be stable than to result in significant morbidity and mortality. However, they observed, “expansion of hemorrhage on any individual case is variable and unpredictable, and it is important to operate at high sensitivities since many patients are taking aspirin or other antiplatelet agents or anticoagulants or may be administered fibrinolytics in the setting of acute stroke.”
PatchFCN is now being applied at UCSF in a study of CT scans from trauma centers across the U.S.