MR Image Segmentation based on Local Information Entropy Geodesic Active Contour Model

Jianwei Zhang, Xiang Ma, Yunjie Chen, Fang Lin, Jin Wang

Abstract

In this paper, an improved Geometric Active Contour model is presented. This model introduced a window function to research the mean information of each pixel's neighbor region, and constructed a novel signed pressure force function based on the entropy to drive the contour towers the boundary. In order to improve the efficiency and stability of the algorithm, this paper adopts two valued level set method to realize the segmentation process. The experimental results show that this method can obtain satisfied result even when the images have intensity inhomogeneity and weak edges.

Introduction

Medical image segmentation—the process of automatically outlining organs or tumors—is a critical yet challenging task. Magnetic Resonance (MR) images frequently suffer from intensity inhomogeneity (bias fields) and weak edges, where tissue transitions are poorly defined. Traditional Active Contour Models (ACMs) generally fall into two categories: edge-based models, which struggle with blurry boundaries and noise, and region-based models, which often fail when intensity varies across the image. This research proposes a model sensitive to local statistical variations to bridge this gap.

Methodology: Local Information Entropy

The core innovation is the Local Information Entropy Signed Pressure Force (LESPF) function. The approach assumes that within a small neighborhood defined by a window function, pixel intensities follow a Gaussian distribution.

Instead of relying solely on intensity, the model calculates the "entropy" (a measure of randomness or information density) of the local region. This entropy serves as a robust signature to distinguish the target object from the background, remaining consistent even when brightness levels shift due to bias fields. The LESPF function acts as a "push-pull" mechanism, driving the contour toward boundaries and stopping precisely when the entropy reaches a specific state. To ensure computational efficiency, the authors utilized a two-valued level set method and replaced complex curvature regularization with a simple Gaussian filter.

Experimental Results and Discussion

The model was evaluated using 20 groups of left ventricle MR images ($110 \times 110$ pixels) characterized by noise, bias fields, and weak boundaries.

Performance Comparison

The proposed method was compared against industry standards, including the Chan-Vese (C-V) model, the standard Geodesic Active Contour (GAC), and the Local Binary Fitting (LBF) model.

Comparison of segmentation results: (a) initial contour, (b) result of comparison model, (c) over-segmentation in traditional model, and (d) proposed entropy-based model result.

Key findings include:

  • Accuracy: The method achieved a Jaccard Similarity (JS) index exceeding 90%, significantly outperforming RGAC and LBF models on difficult images.
  • Robustness: The model effectively ignored artifacts and "labeled lines" that typically cause over-segmentation (where the contour spills into adjacent tissues) in other models.
  • Efficiency: Average iteration time was under 3 seconds, typically requiring fewer than 200 iterations, making it viable for practical use.

Conclusion and Future Work

By prioritizing local entropy over global intensity, the proposed model achieves high accuracy and robustness against the common pitfalls of MRI segmentation. While successful on 2D slices of the left ventricle, future work will focus on extending this approach to 3D volumetric data and further optimizing parameter selection for diverse tissue types.

Summary updated

Abstract

In this paper, an improved Geometric Active Contour model is presented. This model introduced a window function to research the mean information of each pixel's neighbor region, and constructed a novel signed pressure force function based on the entropy to drive the contour towers the boundary. In order to improve the efficiency and stability of the algorithm, this paper adopts two valued level set method to realize the segmentation process. The experimental results show that this method can obtain satisfied result even when the images have intensity inhomogeneity and weak edges.
BibTeX
@article{zhang2014mr,
  author = {Jianwei Zhang and Xiang Ma and Yunjie Chen and Fang Lin and Jin Wang},
  title = {{MR Image Segmentation based on Local Information Entropy Geodesic Active Contour Model}},
  journal = {International Journal of Multimedia and Ubiquitous Engineering},
  year = {2014},
  doi = {10.14257/ijmue.2014.9.4.14},
}
Figure