Preoperative Thoracic Aortic Repair Planning using Deep Learning and 3D Visualization

Master thesis (2024)

Authors

J.A. van Stiphout Mechanical Engineering

Contributors

S. Abo Seada (mentor)
Bardia Arabkhani (mentor)
Q.J. Mank (mentor)
Faculty
Mechanical Engineering, Mechanical Engineering
More Info
expand_more

Published Date

17-07-2024

Language

English

Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Faculty

Mechanical Engineering

Abstract

Introduction: Surgery is needed when the thoracic aorta, the largest artery of the human body, is dilated or the inner wall is ruptured. In preoperative CT imaging, the anatomy can be assessed, and diameters can be measured to detect indications for surgery. A downside of this 3D imaging technique is that its conventional visualization is 2D. By annotating the thoracic aorta in a CT scan and adding together all labeled pixels, a 3D object can be created. The use of 3D visualization for preoperative planning of thoracic aortic repair already showed promising results.

Objective: The objective of this research is to train and test a deep learning model to automatically annotate the thoracic aorta on CT imaging. Based on an automatically reconstructed centerline, the diameter of the aorta can be determined perpendicular to the centerline. All this together will be implemented in a 3D visualization, which will be tested by aortic surgeons as a preoperative aortic repair planning tool.
Methods: A deep learning model was trained by CT scans and corresponding manual annotations of the thoracic aorta. To test the performance, the overlap of manual and automatic segmentations of a test dataset was determined. The created 3D visualization was used by aortic surgeons in the preoperative aortic repair planning of five patient cases. Manual diameter measurements were taken to compare to the automatic diameter measurements. The aortic surgeons who tested the 3D visualization were questioned about the ease of use, usefulness, and their attitude toward working with this software in the future.
Results: The average overlap of the manually and automatically created 3D models was 97.4%. The automatic diameter measurements showed high correspondence when compared to manual measurements done by four aortic surgeons. All surgeons responded positively to the questions concerning usefulness. According to the respondents, the software should be used as a supplement to conventional CT visualization, and not instead of.

Conclusion: The automatically created 3D visualization of the thoracic aorta based on CT imaging was successfully demonstrated. It showed a high ease of use and was of added value in the preoperative assessment of thoracic aortic anatomy. Further validation of the automatic diameter measurements is needed to implement this visualization in daily practice.