Pairwise Active Appearance Model and Its Application to Echocardiography Tracking

Shaohua Kevin Zhou, Jie Shao, Bogdan Georgescu, Dorin ComaniciuMICCAI 2006 · OpenReview

Abstract

We propose a pairwise active appearance model (PAAM) to characterize statistical regularities in shape, appearance, and motion presented by a target that undergoes a series of motion phases, such as the left ventricle in echocardiography. The PAAM depicts the transition in motion phase through a Markov chain and the transition in both shape and appearance through a conditional Gaussian distribution. We learn from a database the joint Gaussian distribution of the shapes and appearances belonging to two consecutive motion phases (i.e., a pair of motion phases), from which we analytically compute the conditional Gaussian distribution. We utilize the PAAM in tracking the left ventricle contour in echocardiography and obtain improved tracking results in terms of localization accuracy when compared with expert-specified contours.

Introduction

Tracking the left ventricle (LV) in echocardiography is a critical task in cardiac imaging, yet it remains notoriously difficult due to speckle noise, signal dropouts, and artifacts from respiratory interference. Traditional methods like Active Shape Models (ASM) and Active Appearance Models (AAM) are primarily designed for static images and often struggle with the rapid shape changes and global motion characteristic of the cardiac cycle. This paper introduces the Pairwise Active Appearance Model (PAAM), a framework that captures shape, appearance, and motion transitions simultaneously to provide robust, frame-by-frame tracking in noisy ultrasound data.

Methodology

The core innovation of the PAAM is its "pairwise" approach, which models the statistical transitions between consecutive motion phases rather than treating each state in isolation.

Motion Phase Modeling

The cardiac cycle is discretized into $P$ phases. A Markov chain is employed to dictate transitions, where the system either remains in the current phase or advances to the next, ensuring temporal consistency.

Markov chain model for motion phase transitions.

Hierarchical Statistical Learning

The model utilizes a hierarchical PCA approach:

  1. Shape PCA: Performed on landmark point vectors.
  2. Appearance PCA: Performed on pixel intensity vectors.
  3. Joint PCA: A combined analysis of shape and appearance coefficients to identify a latent vector controlling both.

By learning the joint Gaussian distribution of shapes and appearances for pairs of consecutive phases, the model analytically derives a conditional Gaussian distribution. This allows the PAAM to predict the most likely configuration in frame $t$ based on the observations from frame $t-1$.

Fusion Tracking

The tracking process uses a fusion approach, combining local optical flow for individual landmarks with the PAAM's global regularization. This ensures that the final tracked contour remains a plausible anatomical shape even when local image evidence is weak.

Experimental Setup

The model was evaluated using a clinical dataset consisting of 400 A4C (apical four-chamber) and 320 A2C (apical two-chamber) sequences, totaling over 20,000 frames. Ground truth was established via manual expert annotations. Performance was measured using 5-fold cross-validation, focusing on localization accuracy and contour smoothness.

Results and Discussion

The PAAM demonstrated significant improvements over baseline methods, such as standard optical flow (SSD) and traditional AAMs. For A2C sequences, the segmental Hausdorff distance (segHD) was reduced from 10.86 pixels to 2.62 pixels.

Statistical variations in shape and appearance across different motion phases (p) and principal components (c).

Beyond accuracy, the model produced much smoother contours. The "surprisal" error—a metric for contour unnaturalness—dropped from 0.32 to 0.0098. Ablation studies revealed that the shape prior is the most critical component for tracking stability, followed by appearance and motion priors. By focusing on transitions, the model maintains robustness even during heavy ultrasound speckle or signal dropouts.

Conclusion

The Pairwise Active Appearance Model provides a robust framework for tracking the left ventricle by integrating motion phase transitions with joint shape and appearance modeling. While the current implementation requires manual initialization, its ability to produce expert-like contours in challenging ultrasound environments marks a significant step toward automated clinical cardiac analysis.

Summary updated

Abstract

We propose a pairwise active appearance model (PAAM) to characterize statistical regularities in shape, appearance, and motion presented by a target that undergoes a series of motion phases, such as the left ventricle in echocardiography. The PAAM depicts the transition in motion phase through a Markov chain and the transition in both shape and appearance through a conditional Gaussian distribution. We learn from a database the joint Gaussian distribution of the shapes and appearances belonging to two consecutive motion phases (i.e., a pair of motion phases), from which we analytically compute the conditional Gaussian distribution. We utilize the PAAM in tracking the left ventricle contour in echocardiography and obtain improved tracking results in terms of localization accuracy when compared with expert-specified contours.
BibTeX
@inproceedings{zhou2006pairwise,
  author = {Shaohua Kevin Zhou and Jie Shao and Bogdan Georgescu and Dorin Comaniciu},
  title = {{Pairwise Active Appearance Model and Its Application to Echocardiography Tracking}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2006},
  url = {https://openreview.net/forum?id=X_SA0UdxDe},
}
Figure