Histological image segmentation using fast mean shift clustering method.

Area of Science:

• Histological image analysis
• Computational pathology
• Digital image processing

Background:

• Accurate color image segmentation is crucial for quantitative histological analysis.
• Histological images present challenges due to staining/illumination variations and high resolution.
• Existing methods struggle to balance segmentation quality with computational cost.

Purpose of the Study:

• To develop an effective and reliable histological image segmentation approach.
• To address the computational expense of traditional Mean Shift clustering.
• To improve the speed of histological image segmentation without sacrificing accuracy.

Main Methods:

• Utilized Mean Shift clustering for histological image segmentation.
• Transformed RGB images to CIE L*a*b* color space, extracting a* and b* components.
• Accelerated Mean Shift by pre-estimating probability density distribution and employing an integral scheme for vector computation.

Main Results:

• The improved Mean Shift method demonstrated segmentation accuracy on par with standard Mean Shift.
• Experimental results on liver fibrosis images showed the improved method's speed is comparable to k-means.
• Achieved high-quality segmentation results with reduced computational cost.

Conclusions:

• An effective and reliable histological image segmentation approach using improved Mean Shift clustering is presented.
• The method significantly enhances computational efficiency through probability density estimation and an integral scheme.
• This approach offers a viable solution for high-quality, fast histological image segmentation.