Congratulations to Alycen Wiacek! Her first-author journal paper entitled CohereNet: A Deep Learning Architecture for Ultrasound Spatial Correlation Estimation and Coherence-Based Beamforming was accepted to IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control. The paper will appear in the journal’s special issue on Deep Learning in Medical Ultrasound – from image formation to image analysis.
This paper presents details of a novel deep neural network (DNN) architecture, named CohereNet, that was trained to estimate spatial correlation functions. The DNN-estimated correlation functions were then used to create short-lag spatial coherence ultrasound images at a faster rate than a CPU approach and with more accuracy than a GPU approach. Results were generalizable across multiple phantoms, in vivo datasets, ultrasound transducers, and ultrasound system manufacturers not included during training. CohereNet has additional potential benefits in low-power DNN-based FPGA implementations of coherence-based beamforming for miniaturized ultrasound imaging systems. In addition, CohereNet has potential utility in other areas of ultrasound imaging that require fundamental cross-correlation calculations, including elastography, speckle tracking, sound speed correction, and other advanced beamforming algorithms, such as minimum variance beamforming.
Citation: A. Wiacek, E. González and M. A. L. Bell, “CohereNet: A Deep Learning Architecture for Ultrasound Spatial Correlation Estimation and Coherence-Based Beamforming,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, accepted March 20, 2020 [pdf]