Image quality improvement in single plane-wave imaging using deep learning

Plane wave (PW) imaging is fast, but limited by poor imaging quality. Coherent PW compounding (CPWC) improves image quality but decrease frame rate. In this study, we propose a modified CycleGAN model that combines a residual attention module with a space-frequency dual-domain discriminator, termed RADD-CycleGAN, to rapidly reconstruct high-quality ultrasound images. To enhance the ability to reconstruct image details, we specially design a process of hybrid dynamic and static channel selection followed by the frequency domain discriminator. The low-quality images are generated by the 3-angle CPWC, while the high-quality images are generated as real images (ground truth) by the 75-angle CPWC. The training set includes unpaired images, whereas the images in the test set are paired to verify the validity and superiority of the proposed model. Finally, we respectively design ablation and comparison experiments to evaluate the model performance. Compared with the basic CycleGAN, our proposed method reaches a better performance, with a 7.8% increase in the peak signal-to-noise ratio and a 22.2% increase in the structural similarity index measure. The experimental results show that our method achieves the best unsupervised reconstruction from low quality images in comparison with several state-of-the-art methods.

Ultrasound imaging has been widely used in clinical diagnosis because of the characteristics of non-radiation, cost effective and real-time interaction. Hand-held ultrasound device has shown great potential in community health care and telemedicine. However, the image quality and frame rate are compromised due to simplification of hand-held hardware which severely limit its applications. The coherence plane-wave compounding imaging (CPWC) technology is usually used to generate high-quality ultrasound images, but compounding dozens of plane waves (PW) with different incident angles will dramatically reduce the frame rate. To get high-quality image while maintaining the frame rate for hand-held ultrasound imaging, in this paper we propose a LG_Unet model, which combines local detail information and global structure information through residual learning. Additionally, the natural image-based transfer learning with cascade training is adopted to alleviate the learning failure and overfitting issues. The experimental results demonstrated that our model can provide high-quality ultrasound images through only 3 PWs, in which the contrast and lateral resolution of the reconstructed images are better than those of the CPWC results from 31 PWs. The proposed method shows the feasibility of generating high-quality images for hand-held ultrasound device using plane wave while maintaining a promising balance between the image quality and frame rate.

Ultrafast Plane Wave Imaging Using Tensor Completion-Based Minimum Variance Algorithm.

Plane-wave imaging (PWI) is a best way to improve the frame rate in ultrasound imaging. However, due to the lack of transmit focus in PWI, the image quality is poor. A novel coherent plane-wave compounding (CPWC) was proposed to handle this issue, but the increased plane waves (PWs) reduced the frame rate. Thus, we combine a novel Double-Stage Delay Multiply and Sum (DS-DMAS) beamformer, which is shown to be better to improve the image quality, with plane-wave compounding to analyze the performance in terms of resolution and contrast. The simulation tests are carried out to show that applying DS-DMAS beamformer can achieve better results even if the number of transmitted PWs are decreased.

Plane wave ultrasound imaging using compressive sensing and minimum variance beamforming

Synthetic aperture focusing based on planar waves also known as coherent plane wave compounding (CPWC) was introduced to reduce the diffraction spreading effect of ultrasound at depths (Song et al. US Patent 6736780, 2004). CPWC has been widely used in shear wave elasticity imaging due to its capability of ultrafast frame rate (Montaldo, et al. IEEE TUFFC, vol. 56, 2009). However, CPWC based on unfocused (plane waves) or defocused Tx beams has an inherent limitation in imaging wider field of view (FOV) due to the use of sampling angle in transmitting steered ultrasound wave. The sampling angle causes an unwanted side-lobe artifact if the sampling interval of angle is not small enough to cancel the artifact.

Ultrasound strain imaging has great importance as tumor or cancerous tissues are much stiffer than normal tissues. Estimations of strain calculating the time delays between consecutive ultrasound echo signals are common approaches in quasi-static elastography. Because of the ill posed nature of strain images, the ultrasound strain estimation still remains challenging. In this paper, we present an iterative strain estimator that uses mutual information between stochastic pre and post compressed radio frequency (RF) signals to define time delays. For each RF signal, a past and future is defined by cutting sample sequences into two parts. The time delay is considered to be the time shift between the cutting moments for which the mutual information between both past vectors and future vectors reaches a minimum. Displacement calculation with this time delay provides effective strain estimation for small strains. A synthetic phantom of tissue with tumor is modeled using the finite element method (FEM) and the pre-/post-compressed RF signals are generated using ultrasound simulation package Field II. Using these simulated RF data displacements and strain of deformation are calculated in MATLAB. The Contrast to Noise Ratio (CNR) and Signal to Noise Ratio (SNR) are found better compared to conventional cross-correlation method. This algorithm can detect inclusions having echogenicity very close to that of the surrounding tissues and hence can be useful for detection of tumors or cancers.

In this paper we present a study for designing realistic ultrasound image simulations from a statistical point of view. Indeed it is extremely important to be able to compute realistic simulated images for validating segmentation or classification methods based on the statistics of ultrasound images. The statistics of the radio frequency (RF) signals is modeled by a distribution called K-RF distribution. The scatterers modeling the medium are uniformly spatially distributed and have scattering coefficients modeled by a K-distribution. The concentration of scatterers and the statistics of their amplitude determine the statistics of the RF signals. We show how scale and shape parameters of the K-RF distribution modeling the RF signal are influenced by concentration, scale and shape parameters of the K distribution modeling the scatterers echogenecity. This study is carried out with the field II simulation program.

Since its inception, ultrasound has been recognized to be one of the most significant technologies in clinical imaging. The rapid and continuous development of ultrasound scanning technologies has added another challenge of rendering relatively large scale 3D volumes with high resolution. Moreover, some applications have required particular high quality ultrasound rendering at interactive frame rate for improved interpretation of the data. Based on the vast computing power of the modern Graphics Processing Units (GPUs), we present an efficient ultrasound reconstruction pipeline that is capable of rendering high quality and high resolution ultrasound images at interactive frame rates for large scale volumes. The reconstruction results of our rendering pipeline have been applied on a real fetal ultrasound data. The performance benchmarks of our pipeline demonstrate its capability of rendering high quality ultrasound images for 10243 volumes with resolutions of 20482 at interactive frame rates using a recent commodity GPU.

Ultrasound Fourier imaging was first initiated by Jian-yu Lu during the 90's. Using this method, one can compute an ultrasound image using a single emission, allowing a very high frame rate (up to 10000 frames per second). The main limitation of this method is the presence of geometrical artifacts, which tend to reduce the image resolution. In this study, we propose to use steered plane waves in reception instead of spherical waves as in [1]. This helps sampling the non-null part of the ultrasound spectrum with finer resolution, while reducing the presence of artifacts. A plane wave is emitted. Back-scattered signals are measured using multiple steered plane waves. Those signals 1D spectrum contain spatial information merged to create the 2D ultrasound image spectrum. Due to axial modulation, this 2D spectrum features 2 symmetric lobes. The image resolution increases when using plane waves instead of spherical waves because the sampling of these lobes is easier. The PSF size is measured both in simulation and experimentally for each method. For conventional imaging technique, this size is 155 by 311µm in axial and transverse directions. Standard Fourier imaging leads to a size of 156 by 279µm. With our spectral imaging, this size is 154µm in axial direction and 252µm in transverse direction. Using steered plane wave in a Fourier imaging framework increases the beamformed image resolution, especially in the transverse direction when compared to other beamforming methods.

Blind super-resolution for handheld ultrasound image: Two-stage degradation based unpaired deep learning.

Tele-intensivists can instruct non-physicians to acquire high-quality ultrasound images

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.

Improvement of low-frequency ultrasonic image quality using a enhanced convolutional neural network

Ultrasound imaging is a widely used and safe medical diagnostic technique, due to its noninvasive nature, low cost, capability of forming real time imaging, and the continuing improvements in image quality. However; the usefulness of ultrasound imaging is degraded by the presence of signal dependant noise known as speckle. It is well-known that speckle is a multiplicative noise that degrades the visual evaluation in ultrasound imaging. In ultrasound (US) imaging, denoising is intended to improve quantitative image analysis techniques. In this paper, a new version of the Non Local (NL-) means filter adapted for US images is proposed based on Similarity function depend on specific characteristics of the variance speckle noise in ultrasound images. The proposed method has been compared with Median, Wavelet, Mean and variance local statistics, Geometric, Anisotropic diffusion filtering, and Non ' local means filter using quantitative parameters. From the visual results and image quality evaluation metrics obtained over real images we can conclude that the modified(NL-) means filter can be successfully used for ultrasound image denoising, and performs better results than all other methods while still retaining the structural details and retains the edges and textures very well while removing speckle noise.