[Paper Review] A dynamic graph-cuts method with integrated multiple feature maps for segmenting kidneys in ultrasound images

Qiang Zheng, Steven L. Warner+2 more|arXiv (Cornell University)|Jun 11, 2017

Medical Image Segmentation Techniques29 references3 citations

TL;DR

This paper proposes a dynamic graph-cuts method that integrates image intensity and Gabor-derived texture features to improve kidney segmentation in ultrasound images. By constructing a localized graph around kidney boundaries and iteratively updating edge weights using regional similarity, the method achieves high accuracy with a mean Dice score of 0.9581 and a mean distance of 1.7166, outperforming state-of-the-art methods (p < 10⁻¹⁹).

ABSTRACT

Purpose: To improve kidney segmentation in clinical ultrasound (US) images, we develop a new graph cuts based method to segment kidney US images by integrating original image intensity information and texture feature maps extracted using Gabor filters. Methods: To handle large appearance variation within kidney images and improve computational efficiency, we build a graph of image pixels close to kidney boundary instead of building a graph of the whole image. To make the kidney segmentation robust to weak boundaries, we adopt localized regional information to measure similarity between image pixels for computing edge weights to build the graph of image pixels. The localized graph is dynamically updated and the GC based segmentation iteratively progresses until convergence. The proposed method has been evaluated and compared with state of the art image segmentation methods based on clinical kidney US images of 85 subjects. We randomly selected US images of 20 subjects as training data for tuning the parameters, and validated the methods based on US images of the remaining 65 subjects. The segmentation results have been quantitatively analyzed using 3 metrics, including Dice Index, Jaccard Index, and Mean Distance. Results: Experiment results demonstrated that the proposed method obtained segmentation results for bilateral kidneys of 65 subjects with average Dice index of 0.9581, Jaccard index of 0.9204, and Mean Distance of 1.7166, better than other methods under comparison (p&lt;10-19, paired Wilcoxon rank sum tests). Conclusions: The proposed method achieved promising performance for segmenting kidneys in US images, better than segmentation methods that built on any single channel of image information. This method will facilitate extraction of kidney characteristics that may predict important clinical outcomes such progression chronic kidney disease.

Motivation & Objective

• To address the challenge of high appearance variation and weak boundaries in kidney ultrasound images.
• To improve segmentation accuracy by combining multiple image features, including intensity and texture.
• To enhance computational efficiency by focusing graph construction only on pixels near kidney boundaries.
• To develop a robust, iterative segmentation framework that dynamically updates graph structure for improved convergence.
• To enable reliable extraction of kidney characteristics for clinical prediction of chronic kidney disease progression.

Proposed method

• Constructs a graph of image pixels localized near the kidney boundary rather than the entire image to improve computational efficiency.
• Integrates original image intensity and Gabor-filtered texture features as multi-channel input for edge weight computation.
• Uses localized regional similarity to compute edge weights, enhancing robustness to weak or noisy boundaries.
• Dynamically updates the graph structure in each iteration based on evolving segmentation results.
• Applies graph cuts optimization iteratively until convergence, refining the segmentation boundary progressively.
• Employs a region-based similarity measure to guide edge weight assignment, improving segmentation stability.

Experimental results

Research questions

• RQ1 Can integrating multiple feature maps (intensity and texture) improve kidney segmentation accuracy in ultrasound images?
• RQ2 How does dynamic graph updating enhance segmentation robustness to weak or ambiguous boundaries?
• RQ3 Does focusing graph construction on boundary-adjacent pixels improve computational efficiency without sacrificing accuracy?
• RQ4 How does the proposed method compare to single-feature or static graph-cuts approaches in clinical kidney US imaging?
• RQ5 To what extent does the method support reliable extraction of kidney morphological features for clinical prognosis?

Key findings

• The proposed method achieved a mean Dice index of 0.9581 across 65 subjects, significantly outperforming all compared methods (p < 10⁻¹⁹).
• The Jaccard index reached 0.9204, indicating strong overlap between segmented and ground-truth kidney regions.
• The mean distance between segmented and manual contours was 1.7166 pixels, reflecting high spatial accuracy.
• The method demonstrated superior robustness to intensity inhomogeneity and weak boundaries due to localized regional similarity modeling.
• Parameter tuning was performed on 20 subjects, and validation on the remaining 65 subjects confirmed consistent high performance.
• The integration of multiple feature maps (intensity + Gabor texture) led to better segmentation than methods relying on single-channel information.