No abstract

No abstract

Deep learning has gained tremendous popularity as a tool for ultrasound beamforming and image reconstruction. In previous work, we trained deep neural networks (DNNs) to estimate the echogenicity of a medium, to improve acoustical and electronic signal-to-noise ratio (SNR) in channel data, and to detect targeted microbubbles nondestructively for real-time ultrasound molecular imaging. Here, we present several advancements to each application. First, we compare the speckle- and noise-reducing performance of DNNs trained with simple linear Field II simulations of photographic images versus that of DNNs trained with full wave finite-difference time domain numerical simulations containing realistic abdominal walls and the resulting image degradation artifacts. We further extend our nondestructive molecular imaging DNN to incorporate spatiotemporal information using an extended simulation study to increase specificity for stationary bound microbubbles and further improve nondestructive molecular imaging.

Linear frequency-modulated (chirp) transmits have been used successfully in the past to increase penetration depth of ultrasound signals in tissue and to improve the signal-to-noise ratio (SNR) in the resulting ultrasound images. However, beamforming chirp signals using delay-and-sum (DAS) can be slow on systems without a GPU. We propose using the chirp scaling algorithm (CSA), originally developed for synthetic aperture radar, as a faster alternative to DAS on CPU that results in similar image quality, especially at larger depths. To perform preliminary comparisons of the beamforming methods, we simulated in FIELD II monostatic synthetic aperture data containing point targets up to 300 mm in depth. The simulation accounted for average signal attenuation in soft tissue of 0.5 dB per MHz per cm. We analyzed the point spread functions and the runtimes of the methods in MATLAB with a single CPU. Beamformed point targets above 80 mm depth had almost identical sidelobe levels and full-width-at-half-maximum between the two methods, while the median runtime for CSA was 9.3 times faster than for DAS. We further extend CSA to multistatic acquisitions with chirp transmits and with curvilinear arrays for more clinical applicability and improved image quality.

The ability to maintain image uniformity and high resolution over large depths is important for several clinical applications of ultrasound, including deep abdominal imaging in patients with high BMI. One way to improve image quality in such cases is to use retrospective transmit focusing, which involves combining received data from different focused transmits to improve image resolution and SNR outside of transmit focal zone. Retrospective transmit is typically accomplished using the delay-and-sum (DAS) beamforming, which can be slow on systems without a GPU. As a faster alternative, we treat beamsummed signals from focused transmits as monostatic synthetic aperture data from virtual sources, and apply a frequency-domain beamformer such as RDA to rapidly refocus an RF image. We demonstrate the concept using FIELD II simulated ultrasound signals from a point target phantom over 200 mm depth. The RDA-refocused image shows a uniform point spread function, and the reduction in full-width at half maximum by a factor of 2.5 compared to the original image from focused transmits at f-number 10. RDA-based refocusing also achieves a speedup by a factor of 14 relative to the DAS-based retrospective beamformer.