Towards Aortic Digital Twins: From Segmentation Accuracy to Geometric Consistency

Marten Johannes Finck, Julius Daum, Niklas Christoph Koser, Sven Tomforde, Soren Pirk

Keywords: Digital Twin, Geometric Validation, Aorta, TAVR

TL;DR: We present a modular pipeline for the automated construction and evaluation of an aortic digital twin from thoracic CT scans.

Registration Requirement: Yes

Abstract: Transcatheter aortic valve replacement (TAVR) requires accurate assessment of patient-specific aortic anatomy for device selection and risk estimation. However, current planning largely relies on expert-driven geometric measurements, while patient-specific digital twins remain underutilized. We present a modular pipeline for automated construction of aortic digital twins from thoracic CT data, combining deep learning–based lumen segmentation, threshold-based calcification extraction, centerline computation, and simulation-ready mesh generation. This work focuses on the comparative evaluation of two segmentation approaches using an extended validation framework beyond conventional overlap-based metrics. Alongside Dice and related measures, we employ a geometry-aware evaluation based on cross-sectional measurements at predefined anatomical keypoints, allowing direct comparison between reconstructed geometries and the imaging data. Our results show that segmentation accuracy alone is insufficient to ensure geometric consistency, revealing deviations in cross-sectional dimensions that may impact downstream applications such as device sizing or hemodynamic simulation. Code will be released upon acceptance.

Reproducibility: Code will be released upon acceptance.

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 44