Breast-lesion Segmentation Combining B-Mode and Elastography Ultrasound

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Radiology

Background:

• Breast ultrasound (BUS) imaging is vital but has low specificity and sensitivity for cancer detection.
• Artifacts in B-mode ultrasound images hinder lesion detection, leading to false positives and unnecessary biopsies.
• Ultrasound elastography offers improved diagnostic accuracy by providing strain information.

Purpose of the Study:

• To validate a novel lesion segmentation framework for breast ultrasound images.
• To integrate B-mode intensity and strain data for enhanced lesion characterization.
• To improve the accuracy of detecting cancerous breast lesions.

Main Methods:

• A novel lesion segmentation framework utilizing Markov Random Field (MRF) and Maximum a Posteriori (MAP) approaches.
• Integration of B-mode ultrasound intensity and ultrasound elastography strain data.
• Validation using a clinical dataset of 33 breast lesion images (14 cancerous, 19 benign) from two hospitals.

Main Results:

• The combined framework significantly improved segmentation results for cancerous lesions.
• The Dice Similarity Coefficient (DSC) increased from 0.49 (B-mode only) to 0.70 when strain data was included.
• The framework demonstrated effectiveness in segmenting challenging lesions with blurred boundaries.

Conclusions:

• Combining B-mode and strain data in a unified framework enhances breast lesion segmentation accuracy.
• The proposed method shows promise for improving cancer detection and reducing unnecessary biopsies in breast ultrasound.
• This approach is particularly beneficial for segmenting difficult-to-identify cancerous lesions.