2D+T Acoustic Boundary Detection in Echocardiography

Miguel Mulet-Parada, J. Alison Noble

TL;DR

This paper introduces a novel phase-based method for detecting acoustic boundaries in 2D+T echocardiography sequences. By extending 1D phase congruency to the spatio-temporal domain, the algorithm effectively reduces speckle noise and provides feature velocity estimates, offering a more robust approach to cardiac image segmentation than traditional methods.

What Problem Does This Paper Address?

Accurate segmentation of cardiac structures in ultrasound (echocardiography) is crucial for clinical applications, especially with the advent of 3D+T imaging. However, ultrasound images are notoriously challenging due to speckle noise and imaging artifacts, which degrade intensity-based edge detection methods. Existing 2D segmentation algorithms, often borrowed from computer vision, perform poorly on this data. While some methods incorporate temporal information, they often ignore the continuity of motion, which can significantly improve reliability. This paper aims to bridge this gap by developing a robust spatio-temporal feature detection method specifically tailored for echocardiography.

Methods

The core of the proposed method is an extension of the phase congruency concept, a feature detection technique that relies on the alignment of Fourier components rather than image intensity gradients.

  • 1D Phase Congruency: The paper starts by reviewing the 1D phase congruency model, which defines features as points where local phase is consistent across different frequency components. This is related to the local energy model, computed using Hilbert transforms or quadrature filter banks (specifically Log-Gabor filters).
  • 2D Feature Asymmetry: To extend to 2D, the authors use a bank of oriented Log-Gabor filters. Feature detection is then based on "feature asymmetry," which measures the difference between odd and even filter responses. This approach is designed to be selective for step-like edges (like cardiac boundaries) and invariant to intensity and amplitude variations.
  • 2D+T Spatio-Temporal Extension: The key innovation is extending the 2D method to the spatio-temporal (2D+T) domain. This involves designing 3D spatio-temporal Log-Gabor filters oriented in various spatio-temporal directions. The algorithm then computes a 2D+T feature asymmetry measure by taking the maximum response across all filter orientations. This temporal component is crucial for distinguishing persistent cardiac boundaries from transient speckle noise.
  • Feature Velocity Estimation: The method also provides feature velocity estimation by recording which spatio-temporal filter yields the strongest response. The inclination of this filter in the spatio-temporal domain is related to the velocity of the detected feature.

Datasets

The paper uses "typical echocardiogram sequences" and "real 2D cardiac sequences" for evaluation. Specific details about the datasets, such as their origin, size, or whether they are publicly available, are not provided beyond these general descriptions. No new benchmarks are introduced.

Key Insights and Ideas

  • Phase Congruency for Ultrasound: The paper highlights that phase-based methods are more robust to speckle noise and intensity variations in ultrasound imaging compared to gradient-based methods.
  • Temporal Continuity is Key: Incorporating temporal information (2D+T) is essential for distinguishing true cardiac boundaries from spurious, flickering speckle artifacts, which are common in 2D ultrasound.
  • Intensity-Amplitude Invariance: The phase-based approach's invariance to intensity and amplitude variations is a significant advantage for echocardiography, where attenuation artifacts make global intensity thresholds problematic.
  • Feature Asymmetry for Edge Detection: The "feature asymmetry" metric effectively identifies step-like features, which are characteristic of cardiac boundaries, while being less sensitive to other image structures.
  • Velocity Estimation from Filter Response: The inclination of the strongest filter response in the spatio-temporal domain provides a direct estimate of feature velocity, offering a novel way to quantify cardiac motion.

Key Results

  • The 2D+T algorithm successfully reduces flickering speckle features compared to the 2D approach, producing smoother and more persistent boundary detections.
  • The method demonstrates good detection of chamber boundaries in typical echocardiogram sequences.
  • The algorithm is shown to be intensity-amplitude invariant, a crucial property for handling variations in ultrasound image quality.
  • Qualitative results show that the 2D+T method can detect structures that are missed by 2D segmentation.
  • The feature velocity estimation provides color-coded maps and arrow visualizations indicating the direction and speed of cardiac motion during systole and diastole.

Limitations and Future Work

  • Fast Cardiac Motion: The paper acknowledges that the current 2D+T algorithm may struggle with very fast cardiac motions (e.g., mitral valve leaflets during opening) if the frame rate is too low. This limitation is attributed to the data sampling rate rather than the algorithm itself.
  • Future Work:
    • Investigating the use of the symmetry measure for speckle tracking and tissue analysis.
  • Developing a higher-level interpretation system (e.g., using a snake framework) to integrate the segmentation and velocity estimation.
  • Extending the approach to volumetric echocardiographic data.
  • Exploring alternative filter designs and interpolation schemes (e.g., tensor product representations, steerable filters) to improve accuracy.
  • Performing quantitative assessment and evaluation against established protocols like Doppler Tissue Imaging.

Summary updated

Abstract

In this paper we address the problem of spatio-temporal acoustic boundary detection in echocardiography. We propose a phase-based feature detection method to be used as the front end to higher-level 2D+T/3D+T reconstruction algorithms. We develop a 2D+T version of this algorithm and illustrate its performance on some typical echocar-diogram sequences. We show how our temporal-based algorithm helps to reduce the number of spurious feature responses due to speckle and provides feature velocity estimates. Further, our approach is intensity-amplitude invariant. This makes it particularly attractive for echocardiographic segmentation, where choosing a single global intensity-based edge threshold is problematic.
BibTeX
@inproceedings{muletparada1998d,
  author = {Miguel Mulet-Parada and J. Alison Noble},
  title = {{2D+T Acoustic Boundary Detection in Echocardiography}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {1998},
  doi = {10.1007/bfb0056268},
}

Loading related content...

Figure