Learned Super Resolution Ultrasound for Improved Breast Lesion Characterization

Breast cancer is the most common malignancy in women. Mammographic findings such as microcalcifications and masses, well morphologic features of masses sonographic scans, are main diagnostic targets for tumor detection. However, improved specificity these imaging modalities required. A leading alternative target neoangiogenesis. When pathological, it contributes to development numerous types tumors, formation metastases. Hence, demonstrating neoangiogenesis by visualization microvasculature may be great importance. Super resolution ultrasound localization microscopy enables at capillary level. Yet, challenges long reconstruction time, dependency on prior knowledge system Point Spread Function (PSF), separability Ultrasound Contrast Agents (UCAs), need addressed translation super-resolution US into clinic. In this work we use a deep neural network architecture that makes effective signal structure address challenges. We present vivo human results three different breast lesions acquired with clinical scanner. By leveraging our trained network, recovered short without PSF knowledge, requiring UCAs. Each recoveries exhibits corresponds known histological structure. This study demonstrates feasibility super resolution, based scanner, increase promotes diagnosis pathologies.