Blind Source Separation for Clutter and Noise Suppression in Ultrasound Imaging: Review for Different Applications

A simplified and accelerated implementation of SVD for filtering ultrafast power Doppler images

BACKGROUND AND OBJECTIVE

Ultrafast Power Doppler (UPD) is a growing ultrasound modality for imaging and diagnosing microvasculature disease. A key element of UPD is using singular value decomposition (SVD) as a highly selective filter for tissue and electronic noise. However, two significant drawbacks of SVD are its computational burden and the complexity of its algorithms. These limitations hinder the development of fast and specific SVD algorithms for UPD imaging. This study introduces power SVD (pSVD), a simplified and accelerated algorithm for filtering tissue and noise in UPD images.

METHODS

pSVD exploits several mathematical properties of SVD specific to UPD images. In particular, pSVD allows the direct computation of blood-related SVD components from the temporal singular vectors. This feature simplifies the expression of SVD while significantly accelerating its computation. After detailing the theory behind pSVD, we evaluate its performances in several in vitro and in vivo experiments and compare it to SVD and randomized SVD (rSVD).

RESULTS

pSVD strongly decreases the running time of SVD (between 5 and 12 times in vivo) without impacting the quality of UPD images. Compared to rSVD, pSVD can be significantly faster (up to 3 times) or slightly slower but gives access to more estimators to isolate tissue subspaces.

CONCLUSION

pSVD is highly valuable for implementing UPD imaging in clinical ultrasound and provides a better understanding of SVD for ultrasound imaging in general.

Ultrasound Signal Processing: From Models to Deep Learning

Medical ultrasound imaging relies heavily on high-quality signal processing to provide reliable and interpretable image reconstructions. Conventionally, reconstruction algorithms have been derived from physical principles. These algorithms rely on assumptions and approximations of the underlying measurement model, limiting image quality in settings where these assumptions break down. Conversely, more sophisticated solutions based on statistical modeling or careful parameter tuning or derived from increased model complexity can be sensitive to different environments. Recently, deep learning-based methods, which are optimized in a data-driven fashion, have gained popularity. These model-agnostic techniques often rely on generic model structures and require vast training data to converge to a robust solution. A relatively new paradigm combines the power of the two: leveraging data-driven deep learning and exploiting domain knowledge. These model-based solutions yield high robustness and require fewer parameters and training data than conventional neural networks. In this work we provide an overview of these techniques from the recent literature and discuss a wide variety of ultrasound applications. We aim to inspire the reader to perform further research in this area and to address the opportunities within the field of ultrasound signal processing. We conclude with a future perspective on model-based deep learning techniques for medical ultrasound.

Independent Component Analysis Filter for Small Vessel Contrast Imaging During Fast Tissue Motion

Suppressing tissue clutter is an essential step in blood flow estimation and visualization, even when using ultrasound contrast agents. Blind source separation (BSS)-based clutter filter for high-framerate ultrasound imaging has been reported to perform better in tissue clutter suppression than the conventional frequency-based wall filter and nonlinear contrast pulsing schemes. The most notable BSS technique, singular value decomposition (SVD) has shown compelling results in cases of slow tissue motion. However, its performance degrades when the tissue motion is faster than the blood flow speed, conditions that are likely to occur when imaging the small vessels, such as in the myocardium. Independent component analysis (ICA) is another BSS technique that has been implemented as a clutter filter in the spatiotemporal domain. Instead, we propose to implement ICA in the spatial domain where motion should have less impact. In this work, we propose a clutter filter with the combination of SVD and ICA to improve the contrast-to-background ratio (CBR) in cases where tissue velocity is significantly faster than the flow speed. In an in vitro study, the range of fast tissue motion velocity was 5-25 mm/s and the range of flow speed was 1-12 mm/s. Our results show that the combination of ICA and SVD yields 7-10 dB higher CBR than SVD alone, especially in the tissue high-velocity range. The improvement is crucial for cardiac imaging where relatively fast myocardial motions are expected.

Film and Video Quality Optimization Using Attention Mechanism-Embedded Lightweight Neural Network Model

In filming, the collected video may be blurred due to camera shake and object movement, causing the target edge to be unclear or deforming the targets. In order to solve these problems and deeply optimize the quality of movie videos, this work proposes a video deblurring (VD) algorithm based on neural network (NN) model and attention mechanism (AM). Based on the scale recurrent network, Haar planar wavelet transform (WT) is introduced to preprocess the video image and to deblur the video image in the wavelet domain. Additionally, the spatial and channel AMs are fused into the overall network framework to improve the feature expression ability. Further, the residual inception spatial-channel attention (RISCA) mechanism is introduced to extract the multiscale feature information from video images. Meanwhile, skip spatial-channel attention (SSCA) accelerates the network training time to achieve a better VD effect. Finally, relevant experiments are designed, factoring in peak signal-to-noise ratio (PSNR) and structural similarity (SSI). The experimental findings corroborate that the proposed Haar and attention video deblurring (HAVD) outperforms multisize network Haar (MSNH) in PSNR and structural similarity (SSIM), improved by 0.10 dB and 0.005, respectively. Therefore, embedding the dual AMs can improve the model performance and optimize the video quality. This work provides technical support for solving the video distortion problems.

Interpretable Machine Learning for Characterization of Focal Liver Lesions by Contrast-Enhanced Ultrasound

This work proposes an interpretable radiomics approach to differentiate between malignant and benign focal liver lesions (FLLs) on contrast-enhanced ultrasound (CEUS). Although CEUS has shown promise for differential FLLs diagnosis, current clinical assessment is performed only by qualitative analysis of the contrast enhancement patterns. Quantitative analysis is often hampered by the unavoidable presence of motion artifacts and by the complex, spatiotemporal nature of liver contrast enhancement, consisting of multiple, overlapping vascular phases. To fully exploit the wealth of information in CEUS, while coping with these challenges, here we propose combining features extracted by the temporal and spatiotemporal analysis in the arterial phase enhancement with spatial features extracted by texture analysis at different time points. Using the extracted features as input, several machine learning classifiers are optimized to achieve semiautomatic FLLs characterization, for which there is no need for motion compensation and the only manual input required is the location of a suspicious lesion. Clinical validation on 87 FLLs from 72 patients at risk for hepatocellular carcinoma (HCC) showed promising performance, achieving a balanced accuracy of 0.84 in the distinction between benign and malignant lesions. Analysis of feature relevance demonstrates that a combination of spatiotemporal and texture features is needed to achieve the best performance. Interpretation of the most relevant features suggests that aspects related to microvascular perfusion and the microvascular architecture, together with the spatial enhancement characteristics at wash-in and peak enhancement, are important to aid the accurate characterization of FLLs.

Plane-Wave Ultrasound Beamforming Through Independent Component Analysis

BACKGROUND AND OBJECTIVE

Beamforming in coherent plane-wave compounding (CPWC) is an essential step in maintaining high resolution, contrast and framerate. Adaptive methods have been designed to achieve this goal by estimating the apodization weights from echo traces acquired by several transducer elements.

METHODS

Herein, we formulate plane-wave beamforming as a blind source separation problem, where the output of each transducer element is considered as a non-independent observation of the field. As such, beamforming can be formulated as the estimation of an independent component out of the observations. We then adapt the independent component analysis (ICA) algorithm to solve this problem and reconstruct the final image.

RESULTS

The proposed method is evaluated on a set of simulations, real phantom, and in vivo data available from the plane-wave imaging challenge in medical ultrasound. Moreover, the results are compared with other well-known adaptive methods.

CONCLUSIONS

Results demonstrate that the proposed method simultaneously improves the resolution and contrast.

Sampled and discretized of short-time Fourier transform and non-negative matrix factorization: the single-channel source separation case

The Short-time Fourier transform (STFT) is a popular time-frequency representation in many source separation problems. In this work, the sampled and discretized version of Discrete Gabor Transform (DGT) is proposed to replace STFT within the single-channel source separation problem of the Non-negative Matrix Factorization (NMF) framework. The result shows that NMF-DGT is better than NMF-STFT according to Signal-to-Interference Ratio (SIR), Signal-to-Artifact Ratio (SAR), and Signal-to-Distortion Ratio (SDR). In the supervised scheme, NMF-DGT has a SIR of 18.60 dB compared to 16.24 dB in NMF-STFT, SAR of 13.77 dB to 13.69 dB, and SDR of 12.45 dB to 11.16 dB. In the unsupervised scheme, NMF-DGT has a SIR of 0.40 dB compared to 0.27 dB by NMF-STFT, SAR of -10.21 dB to -10.36 dB, and SDR of -15.01 dB to -15.23 dB.

Singular Value Decomposition for Removal of Cardiac Interference from Trunk Electromyogram

A new algorithm based on singular value decomposition (SVD) to remove cardiac contamination from trunk electromyography (EMG) is proposed. Its performance is compared to currently available algorithms at different signal-to-noise ratios (SNRs). The algorithm is applied on individual channels. An experimental calibration curve to adjust the number of SVD components to the SNR (0–20 dB) is proposed. A synthetic dataset is generated by the combination of electrocardiography (ECG) and EMG to establish a ground truth reference for validation. The performance is compared with state-of-the-art algorithms: gating, high-pass filtering, template subtraction (TS), and independent component analysis (ICA). Its applicability on real data is investigated in an illustrative diaphragm EMG of a patient with sleep apnea. The SVD-based algorithm outperforms existing methods in reconstructing trunk EMG. It is superior to the others in the time (relative mean squared error < 15%) and frequency (shift in mean frequency < 1 Hz) domains. Its feasibility is proven on diaphragm EMG, which shows a better agreement with the respiratory cycle (correlation coefficient = 0.81, p-value < 0.01) compared with TS and ICA. Its application on real data is promising to non-obtrusively estimate respiratory effort for sleep-related breathing disorders. The algorithm is not limited to the need for additional reference ECG, increasing its applicability in clinical practice.