Interactive breast mass segmentation using a convex active contour model with optimal threshold values.

Area of Science:

• Medical imaging analysis
• Computational intelligence in healthcare

Background:

• Accurate breast mass segmentation is crucial for diagnosis.
• Traditional active contour models require manual thresholding, which is impractical.
• Existing methods struggle with optimal boundary delineation.

Purpose of the Study:

• To develop an optimized mass-specific threshold for convex active contour models.
• To improve the accuracy of breast mass segmentation on digital mammograms.
• To overcome limitations of fixed thresholding and manual tuning.

Main Methods:

• Utilized particle swarm optimization (PSO) to derive an optimal threshold from the mass probability matrix.
• Applied the optimized threshold to a convex active contour model for segmentation.
• Compared the proposed method against Chan-Vese and a published global segmentation model.

Main Results:

• Achieved high Jaccard similarity indices: 0.89±0.07 (vs. Chan-Vese) and 0.88±0.06 (vs. published model).
• Demonstrated minimal mean Euclidean distance between Fourier descriptors: 0.05±0.03 (vs. Chan-Vese) and 0.06±0.04 (vs. published model).
• The optimized threshold improved segmentation accuracy compared to a fixed threshold of 0.5.

Conclusions:

• The proposed method offers an efficient solution for breast mass segmentation.
• It eliminates the need for initial contour placement and re-initialization.
• Achieves optimal segmentation results for all masses, outperforming fixed threshold approaches.