Deep learning-based whole-heart segmentation in 4D contrast-enhanced cardiac CT

Steffen Bruns,Jelmer M Wolterink,Thomas P.W Van Den Boogert,Jurgen H Runge,Berto J Bouma,José P Henriques,Jan Baan,Max A Viergever,R Nils Planken,Ivana Išgum

doi:10.1016/j.compbiomed.2021.105191

Abstract

Automatic cardiac chamber and left ventricular (LV) myocardium segmentation over the cardiac cycle significantly extends the utilization of contrast-enhanced cardiac CT, potentially enabling in-depth assessment of cardiac function. Therefore, we evaluate an automatic method for cardiac chamber and LV myocardium segmentation in 4D cardiac CT.In this study, 4D contrast-enhanced cardiac CT scans of 1509 patients selected for transcatheter aortic valve implantation with 21,605 3D images, were divided into development (N = 12) and test set (N = 1497). 3D convolutional neural networks were trained with end-systolic (ES) and end-diastolic (ED) images. Dice similarity coefficient (DSC) and average symmetric surface distance (ASSD) were computed for 3D segmentations at ES and ED in the development set via cross-validation, and for 2D segmentations in four cardiac phases for 81 test set patients. Segmentation quality in the full test set of 1497 patients was assessed visually on a three-point scale per structure based on estimated overlap with the ground truth.Automatic segmentation resulted in a mean DSC of 0.89 ± 0.10 and ASSD of 1.43 ± 1.45 mm in 12 patients in 3D, and a DSC of 0.89 ± 0.08 and ASSD of 1.86 ± 1.20 mm in 81 patients in 2D. The qualitative evaluation in the whole test set of 1497 patients showed that automatic segmentations were assigned grade 1 (clinically useful) in 98.5%, 92.2%, 83.1%, 96.3%, and 91.6% of cases for LV cavity and myocardium, right ventricle, left atrium, and right atrium.Our automatic method using convolutional neural networks performed clinically useful segmentation across the cardiac cycle in a large set of 4D cardiac CT images, potentially enabling in-depth assessment of cardiac function.

Highlights

Segmentation of the cardiac chambers and left ventricular (LV) myocardium over the whole cardiac cycle enables the assessment of cardiac morphology and function
These results demon strate that automatic segmentation of the LV myocardium and the LV cavity is more accurate in the ED phase than in the ES phase, while segmentation of the RA is more accurate in the ES phase
The results show that the median Dice similarity coefficient (DSC) is always above 0.85, and the median average symmetric surface distance (ASSD) is always below 1.5 mm, with the left atrium showing the highest DSC and lowest ASSD values

Summary

Introduction

Segmentation of the cardiac chambers and left ventricular (LV) myocardium over the whole cardiac cycle enables the assessment of cardiac morphology and function. Cardiac MRI and echocardiography are standardly used to determine cardiac function. For a growing number of applications, 3D contrast-enhanced cardiac CT scans are acquired over the whole cardiac cycle, resulting in 4D cardiac CT images. These have great potential for in-depth cardiac function assessment and could be used to derive the LV ejection fraction (LVEF) Specification Development set. Test set Patients Male Female Age (y) Scanners. Increment (mm) Reconstructed 3D images Number of phases available Interval between phases.

Methods

Results

Conclusion