AI-Driven Image Segmentation for Preoperative Planning in Endovascular Aneurysm Repair (EVAR)
Keywords:
AI segmentation, EVAR planning, abdominal aortic aneurysm, deep learning, medical imaging.Abstract
Endovascular aneurysm repair (EVAR) relies on the accuracy of the preoperative assessment of the cases of abdominal aortic aneurysms (AAA), and the choice of the device and the procedure strategy depend on the quality of measurements of the anatomy. Traditional segmentation of computed tomography angiography (CTA) or magnetic resonance angiography (MRA) images can be tedious and operator-biased, and can create variability that can influence clinical outcomes. This paper presents a proposed AI-based image segmentation system to automate the process of aortic lumen, thrombus, and other morphological characteristics identification and description necessary during EVAR planning. Training The system is trained on manually annotated deep learning architecture projects on standardized ground-truth labeling vascular imaging datasets including variants of U-net and transformer-based segmentation. The workflow extends to include powerful preprocessing, automatic segmentation and extraction of clinically significant parameters including maximal diameter, neck length and vessel angulation. These deliverables are to be compatible with the already available EVAR planning tools. The framework shows the possibility of decreasing planning time, increasing reproducibility, and decision-making accuracy in contrast to the traditional manual or semi-automated workflows. The initial feedback of radiologists and vascular surgeons suggests that the AI-generated segmentations are very much in line with what is expected of a clinician, and the results are reliable across a range of imaging conditions. Also, the proposed system is designed as a modular construct, which allows its extension to other vascular bed territories, real-time intraoperative guidance and integration into 3D printing or virtual simulation tools. The methodology should help to boost clinical uptake and by enhancing explainability and interpretability to make clinicians more trusting.
