Artificial intelligence for automated thoracic aorta diameter measurement using different computed tomography protocols

Abstract

Abstract This study aimed to develop an automated 3-dimensional (3D) segmentation method for measuring the diameter of the thoracic aorta using different computed tomography (CT) protocols. A total of 587 CT scans were retrospectively analysed, and a manual slice-by-slice segmentation of the thoracic aorta was performed by three specialists. The segmented images were used to train convolutional neural network (CNN) models for automated segmentation. The models achieved high accuracy, with an average Dice Score Coefficient (DSC) of 0.8708. Four different methods for thoracic aorta diameter measurement were compared: manual measuring, semi-automatic measuring, automatic measuring using PyRadiomics, and automatic measuring using a made-to-measure algorithm. The results showed that the automatic measuring methods had similar accuracy to the manual and semi-automatic methods. The mean thoracic aorta diameter varied between 3.3 cm and 4.95 cm. These findings demonstrate the feasibility and accuracy of using artificial intelligence algorithms for automated thoracic aorta diameter measurement, which can aid in the assessment and management of aortic diseases.