Semi-automatic breast ultrasound image segmentation based on mean shift and graph cuts

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate tumor segmentation in breast ultrasound (BUS) images is crucial for diagnosis but remains technically challenging.
• Existing methods often require significant manual intervention or lack robustness.

Purpose of the Study:

• To develop and evaluate a novel semi-automatic method for efficient and accurate tumor segmentation in BUS images.
• To improve the reliability and reduce the computational burden of BUS image analysis.

Main Methods:

• A semi-automatic segmentation approach combining Gaussian filtering, histogram equalization, mean shift, and graph cuts.
• Region of Interest (ROI) selection, image downsampling (bicubic interpolation), and contrast enhancement.
• Automatic seed generation for graph cuts, followed by binary segmentation and morphological refinement.

Main Results:

• Achieved a true positive (TP) rate of 91.7%, a false positive (FP) rate of 11.9%, and a similarity (SI) rate of 85.6%.
• Demonstrated a mean processing time of 0.49 ± 0.36 seconds on standard hardware.
• Validated on a dataset of 69 BUS images (38 benign, 31 malignant) from diverse ultrasound scanners.

Conclusions:

• The proposed semi-automatic method offers a promising solution for BUS tumor segmentation, balancing accuracy and efficiency.
• The technique's performance suggests its potential utility in clinical settings for breast cancer diagnosis support.
• Further research could explore integration into real-time diagnostic systems.