Automatic Left Ventricular Cavity Segmentation via Deep Spatial Sequential Network in 4D Computed Tomography Studies

TL;DR

The paper tackles automated left ventricular cavity segmentation in 4D cardiac CT by introducing a spatial-sequential network (SS-Net) that learns unsupervised 3D motion fields between neighboring time-points and a bi-directional learning (BL) module that fuses segmentation results from both chronological and reverse-chronological directions. A deformation-consistency loss and a spatial transformer guide motion estimation, while L_t = G(I_t, φ_{t-1}) + G(I_t, φ_{t+1}) integrates temporal context to refine segmentation. The approach outperforms state-of-the-art methods on 4D-CT (Dice ~0.955, Jaccard ~0.916, HD ~11.5 mm) and demonstrates generalizability to 4D MR data (ACDC), with improved robustness in challenging ES phases and reduced error propagation. The methodology is validated via 6-fold cross-validation on 18-patient CT data and shows potential for extending to other cardiac structures and modalities, offering a practical impact for automated, consistent cardiovascular analysis.

Abstract

Automated segmentation of left ventricular cavity (LVC) in temporal cardiac image sequences (multiple time points) is a fundamental requirement for quantitative analysis of its structural and functional changes. Deep learning based methods for the segmentation of LVC are the state of the art; however, these methods are generally formulated to work on single time points, and fails to exploit the complementary information from the temporal image sequences that can aid in segmentation accuracy and consistency among the images across the time points. Furthermore, these segmentation methods perform poorly in segmenting the end-systole (ES) phase images, where the left ventricle deforms to the smallest irregular shape, and the boundary between the blood chamber and myocardium becomes inconspicuous. To overcome these limitations, we propose a new method to automatically segment temporal cardiac images where we introduce a spatial sequential (SS) network to learn the deformation and motion characteristics of the LVC in an unsupervised manner; these characteristics were then integrated with sequential context information derived from bi-directional learning (BL) where both chronological and reverse-chronological directions of the image sequence were used. Our experimental results on a cardiac computed tomography (CT) dataset demonstrated that our spatial-sequential network with bi-directional learning (SS-BL) method outperformed existing methods for LVC segmentation. Our method was also applied to MRI cardiac dataset and the results demonstrated the generalizability of our method.

Figures (7)

• Figure 1: Comparison of LVC shape changes during cardiac contraction period using 4D-CT. The images represent the LVC from end-diastole (ED – t0) and end-systole (ES – t4) which shows the deformation caused in LVC contraction from blood pumping out during the systole phase. The red circles and blue arrows are used to indicate the inference of papillary muscles and the trabeculae, which is responsible for the boundary of the LVC in ES period becoming blurry.
• Figure 3: Overview of our spatial-sequential network.
• Figure 4: The bi-directional spatiotemporal motion field.
• Figure 5: LVC segmentation results among the individual time-points for all the comparison methods.
• Figure 6: Segmentation results with various deformations under different time intervals.