Presenter Dr. Sergio Grosu of Ludwig Maximilian University of Munich in Germany and colleagues trained two convolutional neural networks (CNNs) to differentiate premalignant and benign colorectal polyps found on CTC. The first CNN was trained using polyp segmentation masks and the 3D CTC image subvolumes containing the individual polyps. The second CNN was only trained on the CTC subvolumes.
The researchers then assessed the performance of both models on an external multicenter test sample from the Cancer Imaging Archive that included 77 polyps in 59 patients. The first CNN achieved an area under the curve (AUC) of 0.83, 66% sensitivity, and 92% specificity, while the second CNN produced an AUC of 0.75, 65% sensitivity, and 79% specificity for differentiating the colorectal polyps.
"As this method did not necessarily require manual polyp segmentation, it has the potential to facilitate the identification of high-risk polyps as an automated second reader," the authors wrote. "CNN-assisted CTC could improve the diagnostic accuracy of CTC in colorectal cancer screening by allowing for a more precise selection of patients who would benefit from endoscopic polypectomy."
Explore the potential for AI in CTC exams by sitting in on this Sunday morning talk.
