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

Quantitative ultrasound using spectral-based techniques, like the backscatter coefficient (BSC), have demonstrated capabilities for tumor characterization and therapy monitoring. The incorporation of an in situ calibration target, that is, a small titanium bead, can provide more consistent BSC estimates. For analyzing tumors, BSC estimation traditionally relies on manual tumor segmentation and calibration bead detection, a time-consuming and skill-dependent task. This study utilizes a U-Net model for automatic BSC estimation by integrating identification of a titanium calibration target embedded in rabbit mammary tumors with automatic segmentation, enabling real-time applications. The U-Net model demonstrated strong segmentation performance, achieving a Dice score of 0.86. Performance metrics demonstrated reliable BSC parameter estimation, with relative errors of 17.87% for effective scatter diameter (ESD) and 9.95% for effective attenuation concentration (EAC) when comparing automated segmentation to manual segmented tumors, highlighting its potential for accurate, real-time tumor diagnostics and therapy monitoring in clinical practice.