Study on screening and diagnosis of aortic dissection based on non-enhanced CT and deep learning

Abstract Objective To develop a machine learning algorithm to detect aortic dissection on non-contrast-enhanced CT and evaluate the diagnostic ability of the algorithm compared with those of radiologists. Methods This study developed a machine learning algorithm using single-center data collected between May 1st, 2022, and April 30th, 2023. Included in the study were 220 patients (110 with AD and 110 without AD). An AD detection algorithm was developed using a 3D full-resolution U-net architecture. We have continuously trained and developed an algorithm based on machine learning to segment the true and false lumens of the aorta and then determine whether there is aortic dissection. The diagnostic capabilities of our algorithm and three radiologists were also compared. Results The developed algorithm achieved an accuracy of 95.8%, a sensitivity of 93.2%, and a specificity of 92.6%. For radiologists, accuracy, sensitivity, and specificity were 88.6%, 90.6%, and 94.2%, respectively. The algorithm's performance was not significantly different from the mean performance of radiologists in terms of accuracy, sensitivity, or specificity. Conclusion The proposed algorithm showed comparable diagnostic performance to radiologists for detecting AD on non-contrast-enhanced CT, which suggests that the proposed algorithm has the potential to reduce misdiagnosis of AD to improve clinical outcomes.Aortic SegmentationAlgorithm development