Automatic acquisition of the four-chamber view for 3D echocardiography

Xiaoping Liu, Xin YangIEICE 2008 · DOI

Abstract

To assist the diagnosis and analysis of heart disease, we introduce a template matching based method to detect the apical four-chamber (A4C) image plane in the three-dimensional echocardiographic (3DE) dataset. Firstly, a typical A4C image is selected as the template. Secondly, a series of cross-sections are extracted from a 3DE image. Finally, by a histogram-based retrieval, the cross-section most similar to the template is retrieved as the A4C image in this 3DE image. With low computational complexity and easy implementation, this method can detect the A4C image plane in 3DE images of different subjects by only one template image.

Introduction

Three-dimensional echocardiography (3DE) provides a wealth of volumetric data, but it presents a significant navigation challenge for clinicians. To diagnose heart disease or measure ventricular function, specific 2D cross-sections are required, most notably the Apical Four-Chamber (A4C) view.

Manually searching through a 3D volume to find this specific plane is time-consuming and prone to inter-operator variability. While previous research attempted to automate this using complex edge detection or massive template libraries, this paper proposes a simpler, more robust method to automatically locate the A4C view using a single template and histogram-based matching.

Methodology

The proposed pipeline consists of three primary stages based on the geometry of ultrasound acquisition:

  1. Coordinate System Establishment: The 3D ultrasound volume is treated as a truncated pyramid. The cardiac apex is assumed to be at the vertex of this pyramid, with the central axis (OZ) representing the heart's long axis.
  2. Image Database Construction: To identify the A4C plane, the system extracts 180 candidate cross-sections by rotating a plane around the OZ axis (from 0° to 179°). This is performed on the end-diastolic volume, where heart structures are easiest to distinguish.
Figure 1: (a) 3D volume coordinate system and (b) extraction of 2D cross-sections by rotating around the OZ axis.
  1. Histogram-Based Template Matching:
    • Region of Interest: Instead of matching the whole image, the system focuses on a central 50x75 pixel rectangular region. This area typically contains the "septum cross," an anatomical landmark that is more consistent across patients than the entire heart.
  2. Feature Extraction: To handle noise and slight misalignments, cumulative histograms are used as the primary feature.
  3. Selection: The "dissimilarity" between a candidate slice and the template is calculated by summing the absolute differences between their cumulative histograms. The slice with the lowest score is selected as the A4C view.

Experimental Setup

The study utilized 28 full-volume datasets acquired from 16 subjects (ranging from 1 month to 12 years old) using a Philips Sonos7500 system. This demographic range provided a rigorous test for the algorithm, accounting for significant variations in heart size, shape, and image intensity.

Key Results and Insights

  • Accuracy: The method correctly identified the A4C image plane in 23 out of 28 datasets, achieving a success rate of 83%. Results were verified by an experienced sonographer.
  • The "Septum Cross" Anchor: A key insight is focusing on the central septum. While heart size varies wildly with age, the relative distribution of the chambers and the septum landmark remains stable.
  • Robustness and Efficiency: By using cumulative histograms rather than direct pixel matching, the algorithm is invariant to rotation and minor translations. Furthermore, the use of a single template image makes the method computationally lightweight and easy to implement in real-time clinical software.

Conclusion and Future Work

This method provides an efficient solution for the automatic detection of the A4C view in 3DE images. However, limitations remain: the algorithm's success depends on initial template quality and assumes the cardiac apex is centered at the ultrasound pyramid's vertex. Future work will expand this approach to include parasternal and subcostal datasets to provide a more comprehensive cardiac examination suite.

Summary updated

Abstract

To assist the diagnosis and analysis of heart disease, we introduce a template matching based method to detect the apical four-chamber (A4C) image plane in the three-dimensional echocardiographic (3DE) dataset. Firstly, a typical A4C image is selected as the template. Secondly, a series of cross-sections are extracted from a 3DE image. Finally, by a histogram-based retrieval, the cross-section most similar to the template is retrieved as the A4C image in this 3DE image. With low computational complexity and easy implementation, this method can detect the A4C image plane in 3DE images of different subjects by only one template image.
BibTeX
@inproceedings{liu2008automatic,
  author = {Xiaoping Liu and Xin Yang},
  title = {{Automatic acquisition of the four-chamber view for 3D echocardiography}},
  booktitle = {IEICE},
  year = {2008},
  doi = {10.1587/elex.5.316},
}
Figure