Lyam Hollis

Nonlinear multiscale regularisation in MR elastography: Towards fine feature mapping

&NA; Fine‐featured elastograms may provide additional information of radiological interest in the context of in vivo elastography. Here a new image processing pipeline called ESP (Elastography Software Pipeline) is developed to create Magnetic Resonance Elastography (MRE) maps of viscoelastic parameters (complex modulus magnitude |G*| and loss angle ø) that preserve fine‐scale information through nonlinear, multi‐scale extensions of typical MRE post‐processing techniques. Methods: A new MRE image processing pipeline was developed that incorporates wavelet‐domain denoising, image‐driven noise estimation, and feature detection. ESP was first validated using simulated data, including viscoelastic Finite Element Method (FEM) simulations, at multiple noise levels. ESP images were compared with MDEV pipeline images, both in the FEM models and in three ten‐subject cohorts of brain, thigh, and liver acquisitions. ESP and MDEV mean values were compared to 2D local frequency estimation (LFE) mean values for the same cohorts as a benchmark. Finally, the proportion of spectral energy at fine frequencies was quantified using the Reduced Energy Ratio (RER) for both ESP and MDEV. Results: Blind estimates of added noise (&sgr;) were within 5.3% ± 2.6% of prescribed, and the same technique estimated &sgr; in the in vivo cohorts at 1.7 ± 0.8%. A 5 × 5 × 5 truncated Gabor filter bank effectively detects local spatial frequencies at wavelengths &lgr; ≤ 10px. For FEM inversions, mean |G*| of hard target, soft target, and background remained within 8% of prescribed up to Symbol and mean ø results were within 10%, excepting hard target ø, which required redrawing around a ring artefact to achieve similar accuracy. Inspection of FEM |G*| images showed some spatial distortion around hard target boundaries and inspection of ø images showed ring artefacts around the same target. For the in vivo cohorts, ESP results showed mean correlation of Symbol with MDEV and liver stiffness estimates within 7% of 2D‐LFE results. Finally, ESP showed statistically significant increase in fine feature spectral energy as measured with RER for both |G*| (Symbol) and ø (Symbol). Conclusion: Information at finer frequencies can be recovered in ESP elastograms in typical experimental conditions, however scatter‐ and boundary‐related artefacts may cause the fine features to have inaccurate values. In in vivo cohorts, ESP delivers an increase in fine feature spectral energy, and better performance with longer wavelengths, than MDEV while showing similar stability and robustness. Symbol. No caption available. Symbol. No caption available. Symbol. No caption available. Symbol. No caption available. HighlightsNew Magnetic Resonance Elastography (MRE) software pipeline incorporating wavelet‐based denoising and feature‐detection techniques.Systematic noise testing with new Finite Element Method (FEM)–based simulations.Results robust to noise and show new levels of detail for MRE elastograms. Graphical abstract Figure. No caption available.

The simulation of magnetic resonance elastography through atherosclerosis

The clinical diagnosis of atherosclerosis via the measurement of stenosis size is widely acknowledged as an imperfect criterion. The vulnerability of an atherosclerotic plaque to rupture is associated with its mechanical properties. The potential to image these mechanical properties using magnetic resonance elastography (MRE) was investigated through synthetic datasets. An image of the steady state wave propagation, equivalent to the first harmonic, can be extracted directly from finite element analysis. Inversion of this displacement data yields a map of the shear modulus, known as an elastogram. The variation of plaque composition, stenosis size, Gaussian noise, filter thresholds and excitation frequency were explored. A decreasing mean shear modulus with an increasing lipid composition was identified through all stenosis sizes. However the inversion algorithm showed sensitivity to parameter variation leading to artefacts which disrupted both the elastograms and quantitative trends. As noise was increased up to a realistic level, the contrast was maintained between the fully fibrous and lipid plaques but lost between the interim compositions. Although incorporating a Butterworth filter improved the performance of the algorithm, restrictive filter thresholds resulted in a reduction of the sensitivity of the algorithm to composition and noise variation. Increasing the excitation frequency improved the techniques ability to image the magnitude of the shear modulus and identify a contrast between compositions. In conclusion, whilst the technique has the potential to image the shear modulus of atherosclerotic plaques, future research will require the integration of a heterogeneous inversion algorithm.

An anatomically unbiased approach for analysis of renal BOLD magnetic resonance images

Oxygenation defects may contribute to renal disease progression, but the chronology of events is difficult to define in vivo without recourse to invasive methodologies. Blood oxygen level-dependent magnetic resonance imaging (BOLD MRI) provides an attractive alternative, but the R2* signal is physiologically complex. Postacquisition data analysis often relies on manual selection of region(s) of interest. This approach excludes from analysis significant quantities of biological information and is subject to selection bias. We present a semiautomated, anatomically unbiased approach to compartmentalize voxels into two quantitatively related clusters. In control F344 rats, low R2* clustering was located predominantly within the cortex and higher R2* clustering within the medulla (70.96 ± 1.48 vs. 79.00 ± 1.50; 3 scans per rat; n = 6; P < 0.01) consistent anatomically with a cortico-medullary oxygen gradient. An intravenous bolus of acetylcholine caused a transient reduction of the R2* signal in both clustered segments (P < 0.01). This was nitric oxide dependent and temporally distinct from the hemodynamic effects of acetylcholine. Rats were then chronically infused with angiotensin II (60 ng/min) and rescanned 3 days later. Clustering demonstrated a disruption of the cortico-medullary gradient, producing less distinctly segmented mean R2* clusters (71.30 ± 2.00 vs. 72.48 ± 1.27; n = 6; NS). The acetylcholine-induced attenuation of the R2* signal was abolished by chronic angiotensin II infusion, consistent with reduced nitric oxide bioavailability. This global map of oxygenation, defined by clustering individual voxels on the basis of quantitative nearness, might be more robust in defining deficits in renal oxygenation than the absolute magnitude of R2* in small, manually selected regions of interest defined exclusively by anatomical nearness.

A fast method to estimate the wall shear stress waveform in arteries

Introduction Ultrasound has been applied to measure vessel diameter and blood flow velocity to compute the wall shear rate (WSR) in arteries. This paper describes a fast technique to assess the WSR waveform using an image of a pulsed Doppler waveform downloaded from a modern clinical ultrasound scanner. Methods A walled vascular phantom has been developed to mimic the physiological condition of brachial arteries, from where measurements were made. A MATLAB program has been developed and used to compute the WSR waveform in a flow phantom from a pulsed Doppler image. The mean WSR obtained from the WSR waveform was compared with the mean WSR derived from the flow rate obtained from a timed collection method. Measurement errors in Doppler velocity estimates from ultrasound scanners were also investigated and used to determine correction factors in WSR calculations. Results For three different flow phantom depths, 9.5,14.5 and 19.5 mm, the mean percentage errors between the true and measured WSR were found to be 4.5% (SD = 4.0), 7.4% (SD = 5.1) and 14.2% (SD = 4.1) respectively. Conclusions The results demonstrated the feasibility of calculating WSR based solely on an image of the Doppler spectrum and arterial diameter measurement, which opens up the possibility of obtaining WSR estimates from generic scanners.

Investigation of Modelling Parameters for Finite Element Analysis of MR Elastography

Introduction Magnetic resonance elastography (MRE) utilizes mechanically induced shear waves to attain material property measurements of in vivo tissue. Finite element analysis (FEA) can be used to replicate the technique in silico to aid in the testing and development of the MRE post-processing software. This study aimed to investigate the influence of modelling parameters upon FEA of MRE.

Finite element analysis to investigate variability of MR elastography in the human thigh

PURPOSE To develop finite element analysis (FEA) of magnetic resonance elastography (MRE) in the human thigh and investigate inter-individual variability of measurement of muscle mechanical properties. METHODS Segmentation was performed on MRI datasets of the human thigh from 5 individuals and FEA models consisting of 12 muscles and surrounding tissue created. The same material properties were applied to each tissue type and a previously developed transient FEA method of simulating MRE using Abaqus was performed at 4 frequencies. Synthetic noise was applied to the simulated data at various levels before inversion was performed using the Elastography Software Pipeline. Maps of material properties were created and visually assessed to determine key features. The coefficient of variation (CoV) was used to assess the variability of measurements in each individual muscle and in the groups of muscles across the subjects. Mean measurements for the set of muscles were ranked in size order and compared with the expected ranking. RESULTS At noise levels of 2% the CoV in measurements of |G*| ranged from 5.3 to 21.9% and from 7.1 to 36.1% for measurements of ϕ in the individual muscles. A positive correlation (R2 value 0.80) was attained when the expected and measured |G*| ranking were compared, whilst a negative correlation (R2 value 0.43) was found for ϕ. CONCLUSIONS Created elastograms demonstrated good definition of muscle structure and were robust to noise. Variability of measurements across the 5 subjects was dramatically lower for |G*| than it was for ϕ. This large variability in ϕ measurements was attributed to artefacts.

Magnetic Resonance Elastography through Atherosclerosis: A Feasibility Study

It is widely acknowledged that assessing the rupture risk of atherosclerotic plaques, via lumen reduction, is an imperfect criterion and that other properties such as those related to biomechanics may be more relevant. This study investigated the hypothesis that magnetic resonance elastography (MRE) can be used to image the elasticity of atherosclerotic plaques with the aim to give a better indication of rupture risk. Atherosclerotic plaques were imaged through a small feasibility data set including stenosed arterial phantoms, healthy volunteers and peripheral artery disease (PAD) patients. Comparison of the healthy volunteer and PAD patient wave displacement images showed differences in noise levels, wave amplitudes and wave propagation through the lumen. However, the change in shear moduli through healthy and diseased areas of the phantoms and in vivo subjects could not be detected. Synthetic modelling of the arterial phantoms, under replicated imaging conditions, suggested that there is scope to improve the results through increased control of the phantom and the inclusion of more realistic blood mimic. The MRE wave displacement appeared highly damped through the lumen of the atherosclerotic PAD data sets when compared to the healthy volunteers. This interesting result indicates that the presences of disease, likely to be calcified plaques, are causing changes in the wave propagation that may be captured using MRE. There is scope to clarify the conclusions in this study by developing the technique, particularly the imaging acquisition parameters and inversion algorithm.

Computational simulations of MR elastography in idealised abdominal aortic aneurysms

Introduction Patient specific modelling (PSM) of abdominal aortic aneurysm (AAA) aims to predict rupture risk by calculating the peak stress acting on the AAA wall using finite element analysis (FEA). It is hypothesised that magnetic resonance elastography (MRE), a non-invasive technique measuring material properties, can improve PSM by allowing integration of patient specific properties into the model. MRE measurements are, however, dependent on the geometry under investigation as well as the material properties. This preliminary study used FEA to investigate the ability of MRE to achieve reproducible measurements of the elastic properties of the thrombus in different sized idealised AAA geometries. Methods Idealised AAA geometries of diameter 50, 60 and 70 mm were created with material properties based on literature values prescribed. FEA was run with frequencies of 50, 100, and 120 Hz induced into the model. Synthetic noise was applied to the models and the ability of a 3-D Butterworth bandpass filter to remove it’s influence was assessed. Results and Discussion In low prescribed shear moduli greatest accuracy was typically achieved at 50 Hz, contrasting with high prescribed shear moduli, where it was achieved at 120 Hz. Variation in measurements across the three AAAs was lowest at 120 Hz with a mean coefficient of variation across all prescribed shear moduli of 9% in contrast to 11 and 18% at 100 and 50 Hz respectively. Bandpass filtering was able to fully recover material property measurements at noise levels of 1 and 2%, but was unable to do so for high prescribed shear modulus values at levels above this. Conclusions Of the frequencies tested here, 120 Hz achieved the most reproducible measurements across the three AAA sizes, though accuracy of measurements at this frequency was compromised in low prescribed shear moduli.