Related Experiment Video
Updated: Oct 6, 2025
Four-Dimensional Computed Tomography-Guided Valve Sizing for Transcatheter Pulmonary Valve Replacement
Published on: January 20, 2022
Deep learning-based whole-heart segmentation in 4D contrast-enhanced cardiac CT
Steffen Bruns1, Jelmer M Wolterink2, Thomas P W van den Boogert3
1Department of Biomedical Engineering and Physics, Amsterdam UMC, University of Amsterdam, Meibergdreef 9, 1105AZ, Amsterdam, the Netherlands; Amsterdam Cardiovascular Sciences, Amsterdam UMC, Meibergdreef 9, 1105 AZ, Amsterdam, the Netherlands.
An automated method accurately segments cardiac chambers and left ventricular myocardium in 4D cardiac CT scans. This deep learning approach enables detailed cardiac function assessment from contrast-enhanced CT images.
Area of Science:
- Medical Imaging
- Cardiovascular Imaging
- Artificial Intelligence in Medicine
Background:
- Accurate segmentation of cardiac structures in 4D cardiac CT is crucial for assessing cardiac function.
- Current segmentation methods can be time-consuming and operator-dependent.
- Contrast-enhanced cardiac CT offers detailed anatomical information, but its full potential for functional assessment requires robust automated segmentation.
Purpose of the Study:
- To evaluate an automatic method for segmenting cardiac chambers and left ventricular (LV) myocardium in 4D contrast-enhanced cardiac CT.
- To assess the performance of a 3D convolutional neural network (CNN) for cardiac segmentation across the cardiac cycle.
- To determine the clinical utility of automated segmentation in a large patient cohort.
Main Methods:
- A 3D CNN was trained on end-systolic (ES) and end-diastolic (ED) images from 12 patients (development set).
- The method was validated using Dice Similarity Coefficient (DSC) and Average Symmetric Surface Distance (ASSD) on 3D and 2D images from a test set of 1497 patients.
- Qualitative assessment of segmentation quality was performed on a three-point scale for clinical usefulness.
Main Results:
- The automated method achieved a mean DSC of 0.89 ± 0.10 for 3D segmentations and 0.89 ± 0.08 for 2D segmentations.
- Mean ASSD was 1.43 ± 1.45 mm (3D) and 1.86 ± 1.20 mm (2D), indicating high accuracy.
- Clinical usefulness (grade 1) was achieved in over 83% of cases for all evaluated cardiac structures, including LV cavity and myocardium, right ventricle, left atrium, and right atrium.
Conclusions:
- The developed CNN-based automatic segmentation method is highly accurate and reliable for 4D cardiac CT.
- The method provides clinically useful segmentations across the cardiac cycle, facilitating in-depth cardiac function assessment.
- This automated approach has the potential to significantly enhance the clinical utility of contrast-enhanced cardiac CT for cardiovascular evaluation.
