Mammography segmentation with maximum likelihood active contours.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Accurate segmentation of suspicious lesions in digital mammograms is crucial for early breast cancer detection.
• Existing segmentation methods face challenges in achieving high accuracy and robustness.

Purpose of the Study:

• To introduce a novel computer-aided segmentation approach for suspicious lesions in digital mammograms.
• To evaluate the performance and accuracy of the proposed method against established techniques.

Main Methods:

• Development of a maximum likelihood active contour model using level sets (MLACMLS).
• Utilizing the Gamma distribution to model and estimate intensity parameters for lesion and background regions.
• Evaluation on real mammographic images and comparison with adaptive level set-based segmentation method (ALSSM) and spiculation segmentation using level sets (SSLS).

Main Results:

• MLACMLS achieved a segmentation accuracy of 86.85%, outperforming ALSSM (74.32%) and SSLS (57.11%).
• Qualitative comparisons showed superior performance against the Active Contour Without Edge (ACWOE) method.
• Demonstrated the suitability of maximum likelihood as an objective function and the algorithm's robustness to seed point selection.

Conclusions:

• The proposed MLACMLS method offers a significant advancement in digital mammogram segmentation accuracy.
• This approach shows potential for improving computer-aided diagnosis systems for breast cancer detection.
• The robustness and high accuracy make MLACMLS a promising tool for clinical applications.