Accuracy of fully automated iliac artery tortuosity measurements in patients with abdominal aortic aneurysm using deep learning.
Authors
Affiliations (7)
Affiliations (7)
- Department of Vascular and Endovascular Surgery, University Hospital Heidelberg, Im Neuenheimer Feld 420, 69120 Heidelberg, Germany. Electronic address: [email protected].
- Department of Vascular Surgery, RWTH Aachen, Templergraben 55, 52062 Aachen, Germany. Electronic address: [email protected].
- Department of Vascular and Endovascular Surgery, University Hospital Heidelberg, Im Neuenheimer Feld 420, 69120 Heidelberg, Germany. Electronic address: [email protected].
- Department of Vascular and Endovascular Surgery, University Hospital Heidelberg, Im Neuenheimer Feld 420, 69120 Heidelberg, Germany. Electronic address: [email protected].
- Clinic for Diagnostic and Interventional Radiology, University Hospital Heidelberg, Im Neuenheimer Feld 420, 69120 Heidelberg, Germany. Electronic address: [email protected].
- Department of Vascular and Endovascular Surgery, University Hospital Heidelberg, Im Neuenheimer Feld 420, 69120 Heidelberg, Germany. Electronic address: [email protected].
- Department of Vascular and Endovascular Surgery, University Hospital Heidelberg, Im Neuenheimer Feld 420, 69120 Heidelberg, Germany. Electronic address: [email protected].
Abstract
To externally validate the accuracy of deep learning-based iliac artery tortuosity assessment (PRAEVAorta 2, Bordeaux, France) in computed tomography angiography of patients with abdominal aortic aneurysms. Retrospective, multi-center validation study. Data from the ZEnith alPHa for aneurYsm Repair Registry (ZEPHYR) were used. Tortuosity measurements from PRAEVAorta 2 were benchmarked against manual reference measurements from an internationally recognized core laboratory (Syntactx, New York, USA). All PRAEVAorta 2 reports were visually inspected by two observers for plausibility of the segmentation of the iliac arteries. Scans with obvious segmentation errors were excluded. Qualitative and quantitative analysis of potential risk factors for segmentation failures were conducted. The core laboratory performed manual tortuosity measurements using standard multiplanar reconstructions. In total, 270 CTA scans from the ZEPHYR database fulfilled the inclusion criteria and were available for analysis. 21.1 % of scans demonstrated obvious segmentation errors upon initial review and were excluded from this analysis. Most common risk factors for segmentation failure were internal iliac artery occlusions and overlapping aneurysms in the distal infrarenal aorta and proximal common iliac arteries. In the remaining 213 scans, the median iliac tortuosity measured by the core laboratory was 1.34 (IQR 1.23-1.45) while PRAEVAorta 2 reported a median tortuosity of 1.25 (IQR 1.19-1.33). The analysis showed an excellent Pearson correlation of 0.95 (95 % CI: 0.94-0.96) with a moderate to good Intraclass Correlation (ICC) of 0.73 (95 % CI: 0.69-0.77) and a bias of -0.09 in Bland-Altman analysis. PRAEVAorta 2 successfully provided fully automated measurements of the iliac artery tortuosity in most cases, highlighting its potential as a valuable tool for both research and clinical applications. However, a critical review of automatically provided reports by an experienced observer remains essential.