Computer-based margin analysis of breast sonography for differentiating malignant and benign masses

Area of Science:

• Radiology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Distinguishing between malignant and benign solid breast masses is crucial for effective patient management.
• Sonographic imaging is a primary tool for breast mass evaluation, but differentiating subtle features can be challenging.

Purpose of the Study:

• To assess the utility of quantitative margin features in sonographic images for computer-aided diagnosis (CAD) of breast masses.
• To evaluate the association of margin sharpness, echogenicity, and angular variation with malignancy.

Main Methods:

• Quantitative analysis of margin sharpness, echogenicity, and angular variation in 58 biopsy-proven breast masses (38 benign, 20 malignant).
• Logistic regression and receiver operating characteristic (ROC) analysis were used to evaluate feature performance and malignancy probability.
• A 3-feature logistic regression model incorporating age, margin echogenicity, and angular variation was developed.

Main Results:

• Significant differences in margin sharpness, echogenicity, and angular variation were observed between malignant and benign masses (P < .03).
• Malignant masses exhibited less distinct tumor-tissue margins compared to benign masses.
• The 3-feature logistic regression model achieved an area under the ROC curve of 0.87 ± 0.05, indicating good diagnostic performance.

Conclusions:

• Quantitative margin features are reliable for measuring margin distinctiveness in breast masses.
• These quantitative features, when combined with logistic regression, show promise for improving CAD of solid breast lesions.