Deep learning-based tool distinguished between malignancy, controls with 74.7 percent sensitivity for malignancies smaller than 2 cm
WEDNESDAY, Sept. 21, 2022 (HealthDay News) — A deep learning (DL)-based tool may help detect pancreatic cancer on computed tomography (CT) scans, according to a study published online Sept. 13 in Radiology.
Po-Ting Chen, M.D., from the National Taiwan University College of Medicine in Taipei, and colleagues developed and validated a DL-based tool for detecting pancreatic cancer on CT. Contrast-enhanced CT studies in 546 patients diagnosed with pancreatic cancer between January 2006 and July 2018 were retrospectively collected and compared to CT studies of 733 controls with a normal pancreas obtained between January 2004 and December 2019. A segmentation convolutional neural network (CNN) and a classifier ensembling five CNNs was developed and validated in the internal test set and a nationwide real-world validation set. The McNemar test was used to compare the sensitivities of the computer-aided detection (CAD) tool and radiologist interpretation.
The researchers found that the DL tool achieved 89.9 and 95.9 percent sensitivity and specificity, respectively, in the internal test set, with an area under the receiver operating characteristic curve (AUC) of 0.96 and with no significant difference in sensitivity compared with the original radiologist report (96.1 percent). The DL tool distinguished between CT malignant and control studies with 89.7 percent sensitivity and 92.8 percent specificity in a test set of 1,473 real-world CT studies (AUC, 0.95); for malignancies smaller than 2 cm, sensitivity was 74.7 percent.
“The CAD tool may be a useful supplement for radiologists to enhance detection of pancreatic cancer,” the authors write.
One author disclosed financial ties to NVIDIA.
Editorial (subscription or payment may be required)
Copyright © 2022 HealthDay. All rights reserved.
