Abstract:Purpose: Cardiac computed tomography (CT) is widely used in clinical diagnosis of cardiovascular diseases. Whole heart segmentation (WHS) plays a vital role in developing new clinical applications of cardiac CT. However, the shape and appearance of the heart can vary greatly across different scans, making the automatic segmentation particularly challenging. The objective of this work is to develop and evaluate a multiatlas segmentation (MAS) scheme using a new atlas ranking and selection algorithm for automatic WHS of CT data. Research on different MAS strategies and their influence on WHS performance are limited. This work provides a detailed comparison study evaluating the impacts of label fusion, atlas ranking, and sizes of the atlas database on the segmentation performance.Methods: Atlases in a database were registered to the target image using a hierarchical registration scheme specifically designed for cardiac images. A subset of the atlases were selected for label fusion, according to the authors' proposed atlas ranking criterion which evaluated the performance of each atlas by computing the conditional entropy of the target image given the propagated atlas labeling. Joint label fusion was used to combine multiple label estimates to obtain the final segmentation. The authors used 30 clinical cardiac CT angiography (CTA) images to evaluate the proposed MAS scheme and to investigate different segmentation strategies.Results: The mean WHS Dice score of the proposed MAS method was 0.918+/-0.021, and the mean runtime for one case was 13.2 min on a workstation. This MAS scheme using joint label fusion generated significantly better Dice scores than the other label fusion strategies, including majority voting (0.901+/-0.276, p < 0.01), locally weighted voting (0.905+/-0.0247, p < 0.01), and probabilistic patch-based fusion (0.909+/-0.0249, p < 0.01). In the atlas ranking study, the proposed criterion based on conditional entropy yielded a performance curve with higher WHS Dice scores compared to the conventional schemes (p < 0.03). In the atlas database study, the authors showed that the MAS using larger atlas databases generated better performance curves than the MAS using smaller ones, indicating larger atlas databases could produce more accurate segmentation.Conclusions: The authors have developed a new MAS framework for automatic WHS of CTA and investigated alternative implementations of MAS. With the proposed atlas ranking algorithm and joint label fusion, the MAS scheme is able to generate accurate segmentation within practically acceptable computation time. This method can be useful for the development of new clinical applications of cardiac CT. (C) 2015 American Association of Physicists in Medicine.

Multi-Scale Patch and Multi-Modality Atlases for Whole Heart Segmentation of Mri

Multiatlas Whole Heart Segmentation Of Ct Data Using Conditional Entropy For Atlas Ranking And Selection

Segmentation of Cardiac Tissues and Organs for CCTA Images Based on a Deep Learning Model

A Registration and Atlas Propagation Based Framework for Automatic Whole Heart Segmentation of CT Volumes

Whole Heart Segmentation of Cardiac MRI Using Multiple Path Propagation Strategy.

Multi-residual 2D network integrating spatial correlation for whole heart segmentation

A cascaded framework with cross-modality transfer learning for whole heart segmentation

Preserving Cardiac Integrity: A Topology-Infused Approach to Whole Heart Segmentation

A Registration-Based Propagation Framework for Automatic Whole Heart Segmentation of Cardiac MRI

Few-Shot Learning by a Cascaded Framework with Shape-Constrained Pseudo Label Assessment for Whole Heart Segmentation.

Multi-atlas spectral PatchMatch: application to cardiac image segmentation.

MSMHSA-DeepLab V3+: An Effective Multi-Scale, Multi-Head Self-Attention Network for Dual-Modality Cardiac Medical Image Segmentation

An Atlas-Based Segmentation Propagation Framework Using Locally Affine Registration - Application to Automatic Whole Heart Segmentation

Automatic Whole Heart Segmentation Based on Atlas Propagation with A Priori Anatomical Information

Improved 3D Whole Heart Geometry from Sparse CMR Slices

Multi-Planar Deep Segmentation Networks for Cardiac Substructures from MRI and CT

Regional perception and multi-scale feature fusion network for cardiac segmentation

Automatic Segmentation of the Full Heart in Cardiac Computed Tomography Images Using A Haar Classifier and A Statistical Model

An Improved Combination of Faster R-CNN and U-Net Network for Accurate Multi-Modality Whole Heart Segmentation

Evaluation of algorithms for Multi-Modality Whole Heart Segmentation: An open-access grand challenge

Multi-atlas Propagation Based Left Atrium Segmentation Coupled with Super-Voxel Based Pulmonary Veins Delineation in Late Gadolinium-Enhanced Cardiac MRI.