A Nonparametric Shape Prior Constrained Active Contour Model for Segmentation of Coronaries in CTA Images

Yin Wang, Han JiangComputational 2014 · DOI

TL;DR

This paper introduces a novel active contour model for segmenting coronary arteries in CT angiography (CTA) images. It addresses challenges in low-contrast regions and potential segmentation leakage by incorporating an adaptive scale selection mechanism and a nonparametric shape prior, leading to more accurate and detailed vessel extraction.

What Problem Does This Paper Address?

Accurate segmentation of coronary arteries in CTA images is crucial for diagnosing vascular diseases. However, this task is challenging due to the complex geometry of vessels, pathological lesions that can create artifacts, and variations in image features across modalities. Existing methods often struggle with:

  • Low-contrast regions: Where vessel boundaries are not clearly defined by intensity differences alone.
  • Segmentation leakage: Where the segmentation algorithm incorrectly expands beyond the actual vessel boundaries, especially when relying solely on intensity information.
  • Interpatient variability: The significant differences in vessel shapes and sizes between individuals make it difficult to rely on predefined shape models.

This paper aims to overcome these limitations by developing a more robust and accurate segmentation method.

Methods

The proposed method builds upon the active contour (snake) framework, which iteratively deforms a contour to minimize an energy function. The key innovations are:

  • Adaptive Scale Selection Scheme: Instead of using a fixed window size for analyzing local image features, this scheme dynamically adjusts the window size based on the global histogram distribution of the CTA image. This allows the model to be more sensitive to weak edges in low-contrast areas (smaller window) and more robust to noise in high-contrast areas (larger window).
  • Nonparametric Shape Constraint: To prevent leakage and ensure the segmented structures maintain a tubular shape, a nonparametric shape descriptor is introduced. This descriptor analyzes the intensity profile along circular paths around points on the active contour. It leverages the observation that vascular structures have a more concentrated intensity distribution along these profiles compared to non-vascular regions. This constraint penalizes deviations from a tubular shape, effectively guiding the contour away from erroneous expansions.

The overall active contour energy functional combines intensity-based terms (derived from a Gaussian mixture model of image intensities) with the adaptive scale selection and the nonparametric shape constraint.

Datasets

The paper evaluates its method on:

  • Synthetic datasets: Used for initial validation and demonstration of the core concepts.
  • Clinical datasets: Specifically, twelve coronary CT volumes acquired from human subjects. The ground truth for these datasets was established through manual delineation by trained biomedical students. The paper focuses on evaluating the segmentation of three major coronary artery branches.

Key Insights and Ideas

  • Adaptive scale is key for low contrast: The realization that a fixed window size is insufficient for varying image contrasts and that an adaptive approach, informed by global image statistics, significantly improves performance in challenging regions.
  • Shape matters beyond intensity: The insight that intensity alone is not always discriminative and that analyzing the shape of intensity profiles (e.g., along a circle) provides a powerful, nonparametric cue for identifying vascular structures and preventing leakage.
  • Nonparametric shape priors are more flexible: Moving away from rigid, predefined shape models towards a data-driven, nonparametric approach that can adapt to the inherent variability in human anatomy.

Key Results

  • Improved Segmentation Accuracy: The proposed method demonstrated superior performance compared to a localized Chan-Vese (CV) model on both synthetic and clinical datasets.
  • Reduced Leakage: The nonparametric shape constraint effectively prevented the contour from leaking into surrounding tissues, a common issue with intensity-based methods.
  • Enhanced Detail Extraction: The adaptive scale selection allowed for the extraction of more detailed coronary vessel segments, including distal branches.
  • Subvoxel Accuracy: Results on clinical datasets indicated the capability of extracting coronary vessels with subvoxel accuracy, as measured by the Hausdorff distance.
  • Quantitative Metrics: Bar plots show the proposed method achieving higher True Positive (TP) rates and Overlapping Measurement (OM) rates, and lower Hausdorff distances compared to the localized CV model. While the False Positive (FP) rate was slightly higher, suggesting a tendency to oversegment, the overall performance was significantly improved.

Limitations and Future Work

  • Tendency to Oversegment: The authors acknowledge that the proposed method exhibits a slightly higher False Positive (FP) rate, indicating a tendency to oversegment vessels.
  • Future Work: The authors propose extending their research to identify and segment soft plaques within coronary arteries. This is identified as a more challenging task due to the lack of distinct intensity features for plaques.

Summary updated

Abstract

We present a nonparametric shape constrained algorithm for segmentation of coronary arteries in computed tomography images within the framework of active contours. An adaptive scale selection scheme, based on the global histogram information of the image data, is employed to determine the appropriate window size for each point on the active contour, which improves the performance of the active contour model in the low contrast local image regions. The possible leakage, which cannot be identified by using intensity features alone, is reduced through the application of the proposed shape constraint, where the shape of circular sampled intensity profile is used to evaluate the likelihood of current segmentation being considered vascular structures. Experiments on both synthetic and clinical datasets have demonstrated the efficiency and robustness of the proposed method. The results on clinical datasets have shown that the proposed approach is capable of extracting more detailed coronary vessels with subvoxel accuracy.
BibTeX
@inproceedings{wang2014nonparametric,
  author = {Yin Wang and Han Jiang},
  title = {{A Nonparametric Shape Prior Constrained Active Contour Model for Segmentation of Coronaries in CTA Images}},
  booktitle = {Computational},
  year = {2014},
  doi = {10.1155/2014/302805},
}
Figure