Visual computing of dissected aortae

Aortic disease is a broad term which includes aortic aneurysms, dissections, intramual hematoma and other conditions. The management of these conditions varies from long-term pharmacological treatment to immediate aortic surgery. Although different decisions can be made on specific cases, surgical intervention is advised for patients with an aortic diameter larger than 5.50 cm. These diameters are often not always measured at regular positions, due to different medical approaches. Recent work has suggested eleven specific points along the aorta where to measure the aortic diameters; but the view angle as well as exact delineation of the diameter are subject to the user’s image understanding. In this work, we define and validate a deep learning method to automatically retrieve this information in a standardized fashion. Furthermore, we suggest an approach which, for each of these points, provides an estimation of the original diameter before the onset of the disease. The method is executed in three following steps, i) a convolutional neural network segments the aorta, ii) a second deep neural network detects the location of the eleven points, iii) a third neural network reconstructs the original shape of the aorta, which can be used for growth estimation. The resulting measurements are compared with those obtained from different experts. Furthermore, some of the algorithms here discussed have been made available on the public open science platform Studierfenster (http://studierfenster. tugraz. at/).