Edge-guided neural network for quantitative characteristics extraction of b-mode image

Abstract

The majority of previous studies in ultrasound imaging have focused on performance enhancement of brightness mode ultrasonography (B-mode). However, b-mode ultrasonography shows poor precision in differentiating benign and malignant lesions. In contrast, ultrasound computed tomography (USCT), on the basis of the circular transducer array in general, resolves the prevision issue that b-mode sonography has. However, USCT is not popular in clinical applications due to their usage restriction and reconstruction time. In this paper, we propose a novel approach for extracting quantitative characteristics from b-mode images by employing the tomography method in b-mode ultrasonography. An edge-guided neural network is incorporated to reduce the reconstruction time for real-time applications.