Scholarlink

Increased interest in digital-twin based healthcare has stimulated recent advancements in personalised biventricular modelling from cardiac magnetic resonance (CMR) imaging. However, there remains no publicly available end-to-end pipeline for generating structured meshes of the heart across the entire cardiac cycle. This paper presents a new pipeline and tests it on CMR data contributed from two centres. The proposed pipeline comprises view classification, segmentation of the left and right ventricular chambers and myocardium, contour generation, and model fitting in a fully automated sequence, requiring only an image directory as input. The use of 3D U-Nets was explored, and found to increase the temporal coherence of resulting meshes compared to 2D U-Nets when evaluated on 10 test cases. The pipeline is available to be deployed across various applications, including digital-twin based simulations, statistical shape and motion analyses, and clinical research. The pipeline—including code, models, and documentation—can be accessed at https://github.com/UOA-Heart-Mechanics-Research/biv-me .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Open-Source End-to-End Pipeline for Generating 3D+t Biventricular Meshes from Cardiac Magnetic Resonance Imaging

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Joshua R. Dillon,
  • Charlène Mauger,
  • Debbie Zhao,
  • Yu Deng,
  • Steffen E. Petersen,
  • Andrew D. McCulloch,
  • Alistair A. Young,
  • Martyn P. Nash

摘要

Increased interest in digital-twin based healthcare has stimulated recent advancements in personalised biventricular modelling from cardiac magnetic resonance (CMR) imaging. However, there remains no publicly available end-to-end pipeline for generating structured meshes of the heart across the entire cardiac cycle. This paper presents a new pipeline and tests it on CMR data contributed from two centres. The proposed pipeline comprises view classification, segmentation of the left and right ventricular chambers and myocardium, contour generation, and model fitting in a fully automated sequence, requiring only an image directory as input. The use of 3D U-Nets was explored, and found to increase the temporal coherence of resulting meshes compared to 2D U-Nets when evaluated on 10 test cases. The pipeline is available to be deployed across various applications, including digital-twin based simulations, statistical shape and motion analyses, and clinical research. The pipeline—including code, models, and documentation—can be accessed at https://github.com/UOA-Heart-Mechanics-Research/biv-me .