Toward Real-Time Backscatter Coefficient Estimation Incorporating the U-Net Segmentation and an In Vivo Reference...

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Quantitative Ultrasound

Background:

• Quantitative ultrasound (QUS) techniques, such as backscatter coefficient (BSC) estimation, are valuable for tumor characterization and therapy monitoring.
• In situ calibration targets, like titanium beads, improve BSC consistency.
• Traditional BSC estimation involves manual tumor segmentation and calibration bead detection, which is time-consuming and requires expertise.

Purpose of the Study:

• To develop and validate a U-Net model for automated BSC estimation in rabbit mammary tumors.
• To integrate automatic calibration target identification and tumor segmentation for real-time QUS analysis.
• To assess the accuracy of automated BSC parameter estimation compared to manual methods.

Main Methods:

• A U-Net model was employed for simultaneous identification of a titanium calibration target and segmentation of rabbit mammary tumors.
• The model facilitated automated backscatter coefficient (BSC) estimation.
• Performance was evaluated using Dice scores for segmentation and relative errors for effective scatter diameter (ESD) and effective attenuation concentration (EAC).

Main Results:

• The U-Net model achieved a Dice score of 0.86, indicating strong segmentation performance.
• Automated BSC parameter estimation showed acceptable relative errors: 17.87% for ESD and 9.95% for EAC.
• The automated approach demonstrated reliable performance compared to manual segmentation.

Conclusions:

• The developed U-Net model enables accurate and automated BSC estimation in QUS.
• This automated method holds significant potential for real-time tumor diagnostics and therapy monitoring in clinical settings.
• Integration of automated calibration target detection and tumor segmentation streamlines QUS analysis.