Plane-Wave Fourier-Domain Beamforming with CNN-Assisted Resolution Enhancement




Musti Venkata, Shravanthi

Journal Title

Journal ISSN

Volume Title



Ultrafast plane-wave imaging has become a major medical imaging modality with a tremendous potential to advance ultrasound diagnostics. Plane-wave ultrasound imaging enables data acquisition at very high frame rates with a single insonification but can suffer from degraded image quality. The latter can be improved by using multiple plane-wave pulses emitted at different steering angles, which reduces the frame rate but allows for image-enhancing coherent compounding of multiple angle-specific beamformed datasets. With the huge success of deep learning approaches in ultrasound imaging, another promising alternative is to employ an image-enhancing convolutional neural network (CNN) for beamformed data post-processing. This project explores the use of an efficient CNN to enhance the resolution of ultrasound images. Our objective is to keep the plane-wave (PW) image reconstruction cost low. Hence, this work uses fast Fourier-domain migration (FDM) to process the raw channel data in conjunction with a well-known low-complexity CNN called Efficient Sub-Pixel Convolutional Neural Network (ESPCN) to enhance the beamformed data during image reconstruction. To evaluate our approach, we have used two in-vitro experimental datasets from the PICMUS evaluation framework and compared our results with conventional delay-and-sum (DAS) beamforming used for high-resolution image reconstruction. This report shows that the proposed approach outperformed a more expensive DAS beamformer with a significant improvement in B-mode image resolution assessed in terms of full width at half-maximum (FWHM) values measured on 0.1-mm wire targets.



plane-wave imaging, frequency-wavenumber migration, image upscaling, convolutional neural network