OBJECTIVE This article describes the use of ultrasound measurements of physical strain within carotid atherosclerotic plaques as a measure of instability and the potential for vascular cognitive decline, microemboli, and white matter changes. METHODS Asymptomatic patients with significant (> 60%) carotid artery stenosis were studied for dynamic measures of plaque instability, presence of microemboli, white matter changes, and vascular cognitive decline in comparison with normative controls and premorbid state. RESULTS Although classically asymptomatic, these patients showed vascular cognitive decline. The degree of strain instability measured within the atherosclerotic plaque directly predicted vascular cognitive decline in these patients thought previously to be asymptomatic according to classic criteria. Furthermore, 26% of patients showed microemboli, and patients had twice as much white matter hyperintensity as controls. CONCLUSIONS These data show that physical measures of plaque instability are possible through interpretation of ultrasound strain data during pulsation, which may be more clinically relevant than solely measuring degree of stenosis. The data also highlight the importance of understanding that the definition of symptoms should not be limited to motor, speech, and vision function but underscore the role of vascular cognitive decline in the pathophysiology of carotid atherosclerotic disease. Clinical trial registration no.: NCT02476396 (clinicaltrials.gov).

No abstract

Vulnerability and instability in carotid artery plaque has been assessed based on strain variations using noninvasive ultrasound imaging. We previously demonstrated that carotid plaques with higher strain indices in a region of interest (ROI) correlated to patients with lower cognition, probably due to cerebrovascular emboli arising from these unstable plaques. This work attempts to characterize the strain distribution throughout the entire plaque region instead of being restricted to a single localized ROI. Multiple ROIs are selected within the entire plaque region, based on thresholds determined by the maximum and average strains in the entire plaque, enabling generation of additional relevant strain indices. Ultrasound strain imaging of carotid plaques, was performed on 60 human patients using an 18L6 transducer coupled to a Siemens Acuson S2000 system to acquire radiofrequency data over several cardiac cycles. Patients also underwent a battery of neuropsychological tests under a protocol based on National Institute of Neurological Disorders and Stroke and Canadian Stroke Network guidelines. Correlation of strain indices with composite cognitive index of executive function revealed a negative association relating high strain to poor cognition. Patients grouped into high and low cognition groups were then classified using these additional strain indices. One of our newer indices, namely the average L-1 norm with plaque (AL1NWP) presented with significantly improved correlation with executive function when compared to our previously reported maximum accumulated strain indices (MASI). An optimal combination of three of the new indices generated classifiers of patient cognition with an area under the curve (AUC) of 0.880, 0.921 and 0.905 for all (n=60), symptomatic (n=33) and asymptomatic patients (n=27) whereas classifiers using maximum accumulated strain indices alone provided AUC values of 0.817, 0.815 and 0.813 respectively.

No abstract

This study demonstrates the importance of vascular cognitive decline in atherosclerotic disease. This is a function of the degree of instability of the atherosclerotic plaque more than the presence of stroke symptoms. It further suggests that atherosclerotic vascular cognitive decline need not be inevitable, and may be modified by treating hypertension and removal of the unstable plaque. This highlights the need for continued research on the cognitive effects of cerebrovascular disease and the synergistic benefits of intensive medical and surgical therapy.

Ultrasound strain imaging utilizes radio-frequency (RF) ultrasound echo signals to estimate the relative elasticity of tissue under deformation. Due to the diagnostic value inherent in tissue elasticity, ultrasound strain imaging has found widespread clinical and preclinical applications. Accurate displacement estimation using pre and post-deformation RF signals is a crucial first step to derive high quality strain tensor images. Incorporating regularization into the displacement estimation framework is a commonly employed strategy to improve estimation accuracy and precision. In this work, we propose an adaptive variation of the iterative Bayesian regularization scheme utilizing RF similarity metric signal-to-noise ratio previously proposed by our group. The regularization scheme is incorporated into a 2D multi-level block matching (BM) algorithm for motion estimation. Adaptive nature of our algorithm is attributed to the dynamic variation of iteration number based on the normalized cross-correlation (NCC) function quality and a similarity measure between pre-deformation and motion compensated post-deformation RF signals. The proposed method is validated for either quasi-static and cardiac elastography or strain imaging applications using uniform and inclusion phantoms and canine cardiac deformation simulation models. Performance of adaptive Bayesian regularization was compared to conventional NCC and Bayesian regularization with fixed number of iterations. Results from uniform phantom simulation study show significant improvement in lateral displacement and strain estimation accuracy. For instance, at 1.5% lateral strain in a uniform phantom, Bayesian regularization with five iterations incurred a lateral strain error of 104.49%, which was significantly reduced using our adaptive approach to 27.51% (p < 0.001). Contrast-to-noise (CNR e ) ratios obtained from inclusion phantom indicate improved lesion detectability for both axial and lateral strain images. For instance, at 1.5% lateral strain, Bayesian regularization with five iterations had lateral CNR e of −0.31 dB which was significantly increased using the adaptive approach to 7.42 dB (p < 0.001). Similar results are seen with cardiac deformation modelling with improvement in myocardial strain images. In vivo feasibility was also demonstrated using data from a healthy murine heart. Overall, the proposed method makes Bayesian regularization robust for clinical and preclinical applications.

Carotid plaque segmentation in ultrasound longitudinal B-mode images using deep learning is presented in this work. We report on 101 severely stenotic carotid plaque patients. A standard U-Net is compared with a dilated U-Net architecture in which the dilated convolution layers were used in the bottleneck. Both a fully automatic and a semi-automatic approach with a bounding box was implemented. The performance degradation in plaque segmentation due to errors in the bounding box is quantified. We found that the bounding box significantly improved the performance of the networks with U-Net Dice coefficients of 0.48 for automatic and 0.83 for semi-automatic segmentation of plaque. Similar results were also obtained for the dilated U-Net with Dice coefficients of 0.55 for automatic and 0.84 for semi-automatic when compared to manual segmentations of the same plaque by an experienced sonographer. A 5% error in the bounding box in both dimensions reduced the Dice coefficient to 0.79 and 0.80 for U-Net and dilated U-Net respectively.

A multilevel Lagrangian carotid strain imaging algorithm is analyzed to identify computational bottlenecks for implementation on a graphics processing unit (GPU). Displacement tracking including regularization was found to be the most computationally expensive aspect of this strain imaging algorithm taking about 2.2 h for an entire cardiac cycle. This intensive displacement tracking was essential to obtain Lagrangian strain tensors. However, most of the computational techniques used for displacement tracking are parallelizable, and hence GPU implementation is expected to be beneficial. A new scheme for subsample displacement estimation referred to as a multilevel global peak finder was also developed since the Nelder-Mead simplex optimization technique used in the CPU implementation was not suitable for GPU implementation. GPU optimizations to minimize thread divergence and utilization of shared and texture memories were also implemented. This enables efficient use of the GPU computational hardware and memory bandwidth. Overall, an application speedup of was obtained enabling the algorithm to finish in about 50 s for a cardiac cycle. Last, comparison of GPU and CPU implementations demonstrated no significant difference in the quality of displacement vector and strain tensor estimation with the two implementations up to a 5% interframe deformation. Hence, a GPU implementation is feasible for clinical adoption and opens opportunity for other computationally intensive techniques.