Ultrasound Imaging From Sparse RF Samples Using System Point Spread Functions

Acceleration of reconstruction for compressed sensing based synthetic transmit aperture imaging by using in-phase/quadrature data

Compressed sensing-based synthetic transmit aperture (CS-STA) was previously proposed to recover the full radio-frequency (RF) channel dataset of synthetic transmit aperture (STA) from that of a smaller number of randomly apodized plane wave (PW) transmissions. In this way, the imaging frame rate (FR) and contrast are improved with maintained spatial resolution, compared with those of STA. Because CS-STA reconstruction is repeated for all receive elements and RF samples (with a high sampling frequency), the recovery of STA dataset in RF domain is time-consuming. In the meantime, a large amount of RF data needs to be transferred and stored, resulting in an increase of system complexity and required memory space. In this study, CS-STA is extended to in-phase/quadrature (IQ) domain (with lower sampling frequency) for the recovery of baseband STA IQ dataset to accelerate the CS-STA reconstruction by reducing the amount of data to be processed. More importantly, CS-STA reconstruction using IQ data is of practical importance, as clinical ultrasound systems typically record baseband IQ signal instead of RF signal. Simulations, phantom and in vivo experiments verify the feasibility of CS-STA in IQ domain for the recovery of STA dataset. More specifically, CS-STA using IQ data achieves similar image quality and appreciably improves reconstruction speed (by ∼3 times) compared with that using RF data. These findings demonstrate that IQ-domain CS-STA is capable of relieving the computational and storage burdens, which may facilitate the implementation of CS-STA in practical ultrasound systems.

Self-supervised learning for accelerated 3D high-resolution ultrasound imaging

PURPOSE

Ultrasound (US) imaging has been widely used in diagnosis, image-guided intervention, and therapy, where high-quality three-dimensional (3D) images are highly desired from sparsely acquired two-dimensional (2D) images. This study aims to develop a deep learning-based algorithm to reconstruct high-resolution (HR) 3D US images only reliant on the acquired sparsely distributed 2D images.

METHODS

We propose a self-supervised learning framework using cycle-consistent generative adversarial network (cycleGAN), where two independent cycleGAN models are trained with paired original US images and two sets of low-resolution (LR) US images, respectively. The two sets of LR US images are obtained through down-sampling the original US images along the two axes, respectively. In US imaging, in-plane spatial resolution is generally much higher than through-plane resolution. By learning the mapping from down-sampled in-plane LR images to original HR US images, cycleGAN can generate through-plane HR images from original sparely distributed 2D images. Finally, HR 3D US images are reconstructed by combining the generated 2D images from the two cycleGAN models.

RESULTS

The proposed method was assessed on two different datasets. One is automatic breast ultrasound (ABUS) images from 70 breast cancer patients, the other is collected from 45 prostate cancer patients. By applying a spatial resolution enhancement factor of 3 to the breast cases, our proposed method achieved the mean absolute error (MAE) value of 0.90 ± 0.15, the peak signal-to-noise ratio (PSNR) value of 37.88 ± 0.88 dB, and the visual information fidelity (VIF) value of 0.69 ± 0.01, which significantly outperforms bicubic interpolation. Similar performances have been achieved using the enhancement factor of 5 in these breast cases and using the enhancement factors of 5 and 10 in the prostate cases.

CONCLUSIONS

We have proposed and investigated a new deep learning-based algorithm for reconstructing HR 3D US images from sparely acquired 2D images. Significant improvement on through-plane resolution has been achieved by only using the acquired 2D images without any external atlas images. Its self-supervision capability could accelerate HR US imaging.

Main Uncertainties in the RF Ultrasound Scanning Simulation of the Standard Ultrasound Phantoms

Ultrasound echoscopy technologies are continuously evolving towards new modalities including quantitative parameter imaging, elastography, 3D scanning, and others. The development and analysis of new methods and algorithms require an adequate digital simulation of radiofrequency (RF) signal transformations. The purpose of this paper is the quantitative evaluation of RF signal simulation uncertainties in resolution and contrast reproduction with the model of a phased array transducer. The method is based on three types of standard physical phantoms. Digital 3D models of those phantoms are composed of point scatterers representing the weak backscattering of the background material and stronger backscattering from inclusions. The simulation results of echoscopy with sector scanning transducer by Field II software are compared with the RF output of the Ultrasonix scanner after scanning standard phantoms with 2.5 MHz phased array. The quantitative comparison of axial, lateral, and elevation resolutions have shown uncertainties from 9 to 22% correspondingly. The echoscopy simulation with two densities of scatterers is compared with contrast phantom imaging on the backscattered RF signals and B-scan reconstructed image, showing that the main sources of uncertainties limiting the echoscopy RF signal simulation adequacy are an insufficient knowledge of the scanner and phantom’s parameters. The attempt made for the quantitative evaluation of simulation uncertainties shows both problems and the potential of echoscopy simulation in imaging technology developments. The analysis presented could be interesting for researchers developing quantitative ultrasound imaging and elastography technologies looking for simulated raw RF signals comparable to those obtained from real ultrasonic scanning.

Variational Formulation of Unsupervised Deep Learning for Ultrasound Image Artifact Removal

Recently, deep learning approaches have been successfully used for ultrasound (US) image artifact removal. However, paired high-quality images for supervised training are difficult to obtain in many practical situations. Inspired by the recent theory of unsupervised learning using optimal transport driven CycleGAN (OT-CycleGAN), here, we investigate the applicability of unsupervised deep learning for US artifact removal problems without matched reference data. Two types of OT-CycleGAN approaches are employed: one with the partial knowledge of the image degradation physics and the other with the lack of such knowledge. Various US artifact removal problems are then addressed using the two types of OT-CycleGAN. Experimental results for various unsupervised US artifact removal tasks confirmed that our unsupervised learning method delivers results comparable to supervised learning in many practical applications.

Real-Time Passive Acoustic Mapping Using Sparse Matrix Multiplication

Passive acoustic mapping enables the spatiotemporal monitoring of cavitation with circulating microbubbles during focused ultrasound (FUS)-mediated blood-brain barrier opening. However, the computational load for processing large data sets of cavitation maps or more complex algorithms limit the visualization in real-time for treatment monitoring and adjustment. In this study, we implemented a graphical processing unit (GPU)-accelerated sparse matrix-based beamforming and time exposure acoustics in a neuronavigation-guided ultrasound system for real-time spatiotemporal monitoring of cavitation. The system performance was tested in silico through benchmarking, in vitro using nonhuman primate (NHP) and human skull specimens, and demonstrated in vivo in NHPs. We demonstrated the stability of the cavitation map for integration times longer than 62.5 [Formula: see text]. A compromise between real-time displaying and cavitation map quality obtained from beamformed RF data sets with a size of 2000 ×128 ×30 (axial [Formula: see text]) was achieved for an integration time of [Formula: see text], which required a computational time of 0.27 s (frame rate of 3.7 Hz) and could be displayed in real-time between pulses at PRF = 2 Hz. Our benchmarking tests show that the GPU sparse-matrix algorithm processed the RF data set at a computational rate of [Formula: see text]/pixel/sample, which enables adjusting the frame rate and the integration time as needed. The neuronavigation system with real-time implementation of cavitation mapping facilitated the localization of the cavitation activity and helped to identify distortions due to FUS phase aberration. The in vivo test of the method demonstrated the feasibility of GPU-accelerated sparse matrix computing in a close to a clinical condition, where focus distortions exemplify problems during treatment. These experimental conditions show the need for spatiotemporal monitoring of cavitation with real-time capability that enables the operator to correct or halt the sonication in case substantial aberrations are observed.

Compressed sensing for reduced hardware footprint in medical ultrasound

In this work, a compressed sensing method to reduce hardware complexity of ultrasound imaging systems is proposed and experimentally verified. We provide clinical evaluation of the method with a possible high compression rates (up to 64 RF signals compressed into a single channel on receive) which uses elastic net estimation for decoding stage. This allows a reduction in size and power consumption of the front-end electronics with only a minor loss in image quality. We demonstrate an 8-fold receive channel count reduction with a 3.16 dB and 3.64 dB mean absolute error for gallbladder and kidney images, respectively, as well as 7.4% increase in the contrast-to-noise ratio for kidney images and 0.1% loss in the contrast-to noise ratio for gallbladder images, on average. The proposed method may enable a fully portable ultrasonic device with virtually no loss in image quality as compared to a full size clinical scanner to be constructed.

Adaptive and Compressive Beamforming Using Deep Learning for Medical Ultrasound

In ultrasound (US) imaging, various types of adaptive beamforming techniques have been investigated to improve the resolution and the contrast-to-noise ratio of the delay and sum (DAS) beamformers. Unfortunately, the performance of these adaptive beamforming approaches degrades when the underlying model is not sufficiently accurate and the number of channels decreases. To address this problem, here, we propose a deep-learning-based beamformer to generate significantly improved images over widely varying measurement conditions and channel subsampling patterns. In particular, our deep neural network is designed to directly process full or subsampled radio frequency (RF) data acquired at various subsampling rates and detector configurations so that it can generate high-quality US images using a single beamformer. The origin of such input-dependent adaptivity is also theoretically analyzed. Experimental results using the B-mode focused US confirm the efficacy of the proposed methods.

Compressed Sensing with Gaussian Sampling Kernel for Ultrasound Imaging

Recently, compressed sensing (CS) has been applied to ultrasound imaging for either data reduction or frame rate improvement. However, there are no detailed reports yet on strategies for lateral undersampling of channel data in conventional focused beamforming (CFB) and its recovery exploiting the CS approach. We propose a strategic lateral undersampling approach for channel data using the Gaussian sampling scheme and compare it with a direct extension of the often-used uniform undersampling reported for axial undersampling to the lateral direction and 2-D random sampling reported in the literature. As opposed to the reported 2-D random undersampling, we explore undersampling of channel data in the lateral direction by acquiring radiofrequency data from only a reduced number of chosen receive elements and subjecting these data to further undersampling in the axial direction. The effect of the sampling schemes on CS recovery was studied using data from simulations and experiments for various lateral and axial undersampling rates. The results suggest that CS-recovered data from the Gaussian distribution-based channel data subsampling yielded better recovery and contrast in comparison to those obtained from the often-used uniform distribution-based undersampling. Although 90% of the samples from the original data using the proposed sampling scheme were discarded, the contrast of the CS-recovered image was comparable to that of the reference image. Thus, CS with the proposed Gaussian sampling scheme for channel data subsampling not only reduces the data size significantly, but also strategically uses only a few active receive elements in the process; thus, it can provide an attractive option for the affordable point-of-care ultrasound system.

Compressed Ultrasound Signal Reconstruction Using a Low-Rank and Joint-Sparse Representation Model

With the introduction of very dense sensor arrays in ultrasound (US) imaging, data transfer rate and data storage can become a bottleneck in US system design. To reduce the amount of sampled channel data, we propose a new approach based on the low-rank and joint-sparse model that allows us to exploit the correlations between different US channels and transmissions. With this method, the minimum number of measurements at each channel can be lower than the sparsity in compressive sensing theory. The accuracy of the reconstruction is less dependent on the sparse basis. An optimization algorithm based on the simultaneous direction method of multipliers is proposed to efficiently solve the resulting optimization problem. Results on different data sets with different experimental settings show that the proposed method is better adapted to the US signals and can recover the image with fewer samples (e.g., 10% of the samples) than the existing compressive sensing-based methods, while maintaining reasonable image quality.

High-Volume-Rate 3-D Ultrasound Imaging Based on Synthetic Aperture Sequential Beamforming With Chirp-Coded Excitation

Three-dimensional (3-D) ultrasound imaging is a promising modality for many medical applications. Unfortunately, it generates voluminous data in the front end, making it unattractive for high-volume-rate portable medical applications. We apply synthetic aperture sequential beamforming (SASB) to greatly compress the front-end receive data. Baseline 3-D SASB has a low volume rate, because subapertures fire one by one. In this paper, we propose to increase the volume rate of 3-D SASB without degrading imaging quality through: 1) transmitting and receiving simultaneously with four subapertures and 2) using linear chirps as the excitation waveform to reduce interference. We design four linear chirps that operate on two overlapped frequency bands with chirp pairs in each band having opposite chirp rates. Direct implementation of this firing scheme results in grating lobes. Therefore, we design a sparse array that mitigates the grating lobe levels through optimizing the locations of transducer elements in the bin-based random array. Compared with the baseline 3-D SASB, the proposed method increases the volume rate from 8.56 to 34.2 volumes/s without increasing the front-end computation requirement. Field-II-based cyst simulations show that the proposed method achieves imaging quality comparable with baseline 3-D SASB in both shallow and deep regions.