Erschienen in:
11.07.2022 | Original Article
Deep Learning Model for Predicting the Outcome of Endovascular Abdominal Aortic Aneurysm Repair
verfasst von:
Yonggang Wang, Min Zhou, Yong Ding, Xu Li, Zhenyu Zhou, Zhenyu Shi, Weiguo Fu
Erschienen in:
Indian Journal of Surgery
|
Sonderheft 1/2023
Einloggen, um Zugang zu erhalten
Abstract
The objective of this study is to establish a model for predicting the outcome of endovascular aneurysm repair for abdominal aortic aneurysms using deep learning algorithms. We performed a case series analysis of 493 patients with infra-renal abdominal aortic aneurysm who underwent elective endovascular aneurysm repair procedures between January 2016 and December 2019 in our single center. ITK-SNAP software was used to draw the abdominal aortic aneurysms region of interest. Images were preprocessed and deep learning model was built using MATrix LABoratory. Randomly divided, 80% of the patients were used as the training set, and 20% of the patients were enrolled in the test set. The area under the curve from receiver-operating characteristic curve was used to evaluate the predictive effect of the model. To further understand the prediction process of the deep learning model, visualization techniques were used to analyze the features learned by the model and the response of the convolutional layer to the different outcomes of endovascular aneurysm repair. The mean follow-up was 32.0 months, including 156 patients (31.6%) experiencing endovascular aneurysm repair-related severe adverse events. Number of patients in the training set was 394 (of which 269 had no severe adverse events and 125 had severe adverse events) and the number of patients in test set was 99 (of which 68 had no severe adverse events and 31 had severe adverse events). By training on 92,012–93,925 computed tomography angiography images (n = 315) and validation on 22,269–24,182 computed tomography angiography images (n = 79), the deep learning model finally achieved encouraging predictive performance in the test set (n = 99) with an area under the curve of 0.81 SD0.01, accuracy 0.82 SD0.02, and F1 score 0.87 SD0.02. The visualization techniques improved the model interpretability. The deep learning model could be an efficient adjunctive tool to predict outcomes after endovascular aneurysm repair.