On the uncertainty in the segmentation of ultrasound images reconstructed with the total focusing method

Abstract

Article Highlights This study uses neural networks to probabilistically segment TFM images and investigate uncertainty. The probabilistic U-Net outperforms other models, and high uncertainty is found in the shadow zone, reflections, and small defects. Three different defect types are investigated.

Keywords

• Probabilistic image segmentation
• Deep learning
• Total focusing method
• Ultrasound