Impact of denoising on precision and accuracy of saturation-recovery-based myocardial T1 mapping

STEAM-SASHA: a novel approach for blood- and fat-suppressed native T1 measurement in the right ventricular myocardium

OBJECTIVE

The excellent blood and fat suppression of stimulated echo acquisition mode (STEAM) can be combined with saturation recovery single-shot acquisition (SASHA) in a novel STEAM-SASHA sequence for right ventricular (RV) native T1 mapping.

MATERIALS AND METHODS

STEAM-SASHA splits magnetization preparation over two cardiac cycles, nulling blood signal and allowing fat signal to decay. Breath-hold T1 mapping was performed in a T1 phantom and twice in 10 volunteers using STEAM-SASHA and a modified Look-Locker sequence at peak systole at 3T. T1 was measured in 3 RV regions, the septum and left ventricle (LV).

RESULTS

In phantoms, MOLLI under-estimated while STEAM-SASHA over-estimated T1, on average by 3.0% and 7.0% respectively, although at typical 3T myocardial T1 (T1 > 1200 ms) STEAM-SASHA was more accurate. In volunteers, T1 was higher using STEAM-SASHA than MOLLI in the LV and septum (p = 0.03, p = 0.006, respectively), but lower in RV regions (p > 0.05). Inter-study, inter-observer and intra-observer coefficients of variation in all regions were < 15%. Blood suppression was excellent with STEAM-SASHA and noise floor effects were minimal.

DISCUSSION

STEAM-SASHA provides accurate and reproducible T1 in the RV with excellent blood and fat suppression. STEAM-SASHA has potential to provide new insights into pathological changes in the RV in future studies.

Denoising and uncertainty estimation in parameter mapping with approximate Bayesian deep image priors

PURPOSE

To mitigate the problem of noisy parameter maps with high uncertainties by casting parameter mapping as a denoising task based on Deep Image Priors.

METHODS

We extend the concept of denoising with Deep Image Prior (DIP) into parameter mapping by treating the output of an image-generating network as a parametrization of tissue parameter maps. The method implicitly denoises the parameter mapping process by filtering low-level image features with an untrained convolutional neural network (CNN). Our implementation includes uncertainty estimation from Bernoulli approximate variational inference, implemented with MC dropout, which provides model uncertainty in each voxel of the denoised parameter maps. The method is modular, so the specifics of different applications (e.g., T1 mapping) separate into application-specific signal equation blocks. We evaluate the method on variable flip angle T1 mapping, multi-echo T2 mapping, and apparent diffusion coefficient mapping.

RESULTS

We found that deep image prior adapts successfully to several applications in parameter mapping. In all evaluations, the method produces noise-reduced parameter maps with decreased uncertainty compared to conventional methods. The downsides of the proposed method are the long computational time and the introduction of some bias from the denoising prior.

CONCLUSION

DIP successfully denoise the parameter mapping process and applies to several applications with limited hyperparameter tuning. Further, it is easy to implement since DIP methods do not use network training data. Although time-consuming, uncertainty information from MC dropout makes the method more robust and provides useful information when properly calibrated.

Data-driven optimization of sampling patterns for MR brain T1ρ mapping

PURPOSE

The goal of this study was to apply a fast data-driven optimization algorithm, called bias-accelerated subset selection, for MR brain T1ρ mapping to generate optimized sampling patterns (SPs) for compressed sensing reconstruction of brain 3D-T1ρ MRI.

METHODS

Five healthy volunteers were recruited, and fully sampled Cartesian 3D-T1ρ MRIs were obtained. Variable density (VD) and Poisson disc (PD) undersampling was used as the input to SP optimization process. The reconstruction used 3 compressed sensing methods: spatiotemporal finite differences, low-rank plus sparse with spatial finite differences, and low rank. The performance of images and T1ρ maps using PD-SP and VD-SP and their optimized sampling patterns (PD-OSP and VD-OSP) were compared to the fully sampled reference using normalized root mean square error (NRMSE).

RESULTS

The VD-OSP with spatiotemporal finite differences reconstruction (NRMSE = 0.078) and the PD-OSP with spatiotemporal finite differences reconstruction (NRMSE = 0.079) at the highest acceleration factors (AF = 30) showed the largest improvement compared to the respective nonoptimized SPs (VD NRMSE = 0.087 and PD NRMSE = 0.149). Prospective undersampling was tested at AF = 4, with VD-OSP NRMSE = 0.057 versus PD-OSP NRMSE = 0.060, with optimized sampling performing better that input PD or VD sampling. For brain T1ρ mapping, the VD-OSP with low rank reconstruction for AFs <10 and VD-OSP with spatiotemporal finite differences for AFs >10 perform better.

CONCLUSIONS

The study demonstrated that the appropriate use of data-driven optimized sampling and suitable compressed sensing reconstruction technique can be employed to potentially accelerate 3D T1ρ mapping for brain imaging applications.

Image resampling and discretization effect on the estimate of myocardial radiomic features from T1 and T2 mapping in hypertrophic cardiomyopathy

Radiomics is emerging as a promising and useful tool in cardiac magnetic resonance (CMR) imaging applications. Accordingly, the purpose of this study was to investigate, for the first time, the effect of image resampling/discretization and filtering on radiomic features estimation from quantitative CMR T1 and T2 mapping. Specifically, T1 and T2 maps of 26 patients with hypertrophic cardiomyopathy (HCM) were used to estimate 98 radiomic features for 7 different resampling voxel sizes (at fixed bin width), 9 different bin widths (at fixed resampling voxel size), and 7 different spatial filters (at fixed resampling voxel size/bin width). While we found a remarkable dependence of myocardial radiomic features from T1 and T2 mapping on image filters, many radiomic features showed a limited sensitivity to resampling voxel size/bin width, in terms of intraclass correlation coefficient (> 0.75) and coefficient of variation (< 30%). The estimate of most textural radiomic features showed a linear significant (p < 0.05) correlation with resampling voxel size/bin width. Overall, radiomic features from T2 maps have proven to be less sensitive to image preprocessing than those from T1 maps, especially when varying bin width. Our results might corroborate the potential of radiomics from T1/T2 mapping in HCM and hopefully in other myocardial diseases.

In Vivo Super-Resolution Cardiac Diffusion Tensor MRI: A Feasibility Study

A super-resolution (SR) technique is proposed for imaging myocardial fiber architecture with cardiac magnetic resonance. Images were acquired with a motion-compensated cardiac diffusion tensor imaging (cDTI) sequence. The heart left ventricle was covered with three stacks of thick slices, in short axis, horizontal and vertical long axes orientations, respectively. The three low-resolution stacks (2 × 2 × 8 mm³) were combined into an isotropic volume (2 × 2 × 2 mm³) by a super-resolution reconstruction. For in vivo measurements, each slice was acquired during a breath-hold period. Bulk motion was corrected by optimizing a similarity metric between intensity profiles from all intersecting slices in the dataset. The benefit of the proposed approach was evaluated using a numerical heart phantom, a physical helicoidal phantom with artificial fibers, and six healthy subjects. The SR technique showed improved results compared to the native scans, in terms of image quality and cDTI metrics. In particular, the myocardial helix angle (HA) was more accurately estimated in the physical phantom (HA = 41.5° ± 1.1°, with the ground truth being 42.0°). In vivo, it resulted in a sharper rate of change of HA across the myocardial wall (−0.993°/% ± 0.007°/% against −0.873°/% ± 0.010°/%).

Improved cardiac T1 mapping accuracy and precision with a new hybrid MOLLI and SASHA technique: MOSHA

PURPOSE

To develop and validate a new myocardial T₁ mapping sequence (MOSHA) which is based on a combination of the modified Look-Locker inversion recovery (MOLLI) and the saturation recovery single-shot acquisition (SASHA) sequences.

METHODS

Prior studies have shown that myocardial T₁ mapping by SASHA is more accurate but less precise than MOLLI. A new myocardial T₁ mapping technique (MOSHA) based on single-shot acquisitions is developed by combining the MOLLI and SASHA sequences. Phantom and patient studies on 15 patients (9 males, median age 21 years) were performed to validate and compare MOSHA with the MOLLI and SASHA sequences in terms of accuracy and precision.

RESULTS

In the phantom study, MOSHA was as accurate as SASHA (P-value = 0.88) and as precise as MOLLI (P-value = 0.59). Similar trends were observed in the patient study. Compared to SASHA, MOSHA accuracy was comparable for blood pre-contrast (P-value≥0.10) and post-contrast (P-value≥0.70), and for myocardium pre-contrast (P-value = 0.70) and post-contrast (P-value = 0.09). Compared to MOLLI, MOSHA precision was lower for blood pre-contrast (P-value<0.01) and higher for blood post-contrast (P-value≤0.01), and comparable for myocardium pre-contrast (P-value = 0.24) and post-contrast (P-value = 0.07). Synthetic Extracellular volume fraction (ECV) calculated by MOSHA was more precise than those of SASHA and MOLLI (P-value ≤0.01).

CONCLUSION

In phantom studies and patients, MOSHA has comparable accuracy as SASHA and nearly similar precision as MOLLI for T₁ mapping. Precision of MOSHA was better than MOLLI and SASHA in synthetic ECV measurements. Therefore, it may be a superior choice in clinical practice for a precise and accurate calculation of T₁ and ECV.

Faster 3D saturation-recovery based myocardial T1 mapping using a reduced number of saturation points and denoising

PURPOSE

To accelerate the acquisition of free-breathing 3D saturation-recovery-based (SASHA) myocardial T1 mapping by acquiring fewer saturation points in combination with a post-processing 3D denoising technique to maintain high accuracy and precision.

METHODS

3D SASHA T1 mapping acquires nine T1-weighted images along the saturation recovery curve, resulting in long acquisition times. In this work, we propose to accelerate conventional cardiac T1 mapping by reducing the number of saturation points. High T1 accuracy and low standard deviation (as a surrogate for precision) is maintained by applying a 3D denoising technique to the T1-weighted images prior to pixel-wise T1 fitting. The proposed approach was evaluated on a T1 phantom and 20 healthy subjects, by varying the number of T1-weighted images acquired between three and nine, both prospectively and retrospectively. Following the results from the healthy subjects, three patients with suspected cardiovascular disease were acquired using five T1-weighted images. T1 accuracy and precision was determined for all the acquisitions before and after denoising.

RESULTS

In the T1 phantom, no statistical difference was found in terms of accuracy and precision for the different number of T1-weighted images before or after denoising (P = 0.99 and P = 0.99 for accuracy, P = 0.64 and P = 0.42 for precision, respectively). In vivo, both prospectively and retrospectively, the precision improved considerably with the number of T1-weighted images employed before denoising (P<0.05) but was independent on the number of T1-weighted images after denoising.

CONCLUSION

We demonstrate the feasibility of accelerating 3D SASHA T1 mapping by reducing the number of acquired T1-weighted images in combination with an efficient 3D denoising, without affecting accuracy and precision of T1 values.

Accelerated free-breathing whole-heart 3D T2 mapping with high isotropic resolution

PURPOSE

To enable free-breathing whole-heart 3D T2 mapping with high isotropic resolution in a clinically feasible and predictable scan time. This 3D motion-corrected undersampled signal matched (MUST) T2 map is achieved by combining an undersampled motion-compensated T2 -prepared Cartesian acquisition with a high-order patch-based reconstruction.

METHODS

The 3D MUST-T2 mapping acquisition consists of an electrocardiogram-triggered, T2 -prepared, balanced SSFP sequence with nonselective saturation pulses. Three undersampled T2 -weighted volumes are acquired using a 3D Cartesian variable-density sampling with increasing T2 preparation times. A 2D image-based navigator is used to correct for respiratory motion of the heart and allow 100% scan efficiency. Multicontrast high-dimensionality undersampled patch-based reconstruction is used in concert with dictionary matching to generate 3D T2 maps. The proposed framework was evaluated in simulations, phantom experiments, and in vivo (10 healthy subjects, 2 patients) with 1.5-mm3 isotropic resolution. Three-dimensional MUST-T2 was compared against standard multi-echo spin-echo sequence (phantom) and conventional breath-held single-shot 2D SSFP T2 mapping (in vivo).

RESULTS

Three-dimensional MUST-T2 showed high accuracy in phantom experiments (R2 > 0.99). The precision of T2 values was similar for 3D MUST-T2 and 2D balanced SSFP T2 mapping in vivo (5 ± 1 ms versus 4 ± 2 ms, P = .52). Slightly longer T2 values were observed with 3D MUST-T2 in comparison to 2D balanced SSFP T2 mapping (50.7 ± 2 ms versus 48.2 ± 1 ms, P < .05). Preliminary results in patients demonstrated T2 values in agreement with literature values.

CONCLUSION

The proposed approach enables free-breathing whole-heart 3D T2 mapping with high isotropic resolution in about 8 minutes, achieving accurate and precise T2 quantification of myocardial tissue in a clinically feasible scan time.

Accelerated mono- and biexponential 3D-T1ρ relaxation mapping of knee cartilage using golden angle radial acquisitions and compressed sensing

PURPOSE

To use golden-angle radial sampling and compressed sensing (CS) for accelerating mono- and biexponential 3D spin-lattice relaxation time in the rotating frame (T_1ρ ) mapping of knee cartilage.

METHODS

Golden-angle radial stack-of-stars and Cartesian 3D-T_1ρ -weighted knee cartilage datasets (n = 12) were retrospectively undersampled by acceleration factors (AFs) 2-10. CS-based reconstruction using 8 different sparsifying transforms were compared for mono- and biexponential T_1ρ -mapping of knee cartilage, including spatio-temporal finite differences, wavelets, dictionary from principal component analysis, and exponential decay models, and also low rank and low rank plus sparse models (L+S). Complex-valued fitting was used and Marchenko-Pastur principal component analysis filtering also tested.

RESULTS

Most CS methods performed well for an AF of 2, with relative median normalized absolute deviation below 10% for monoexponential and biexponential mapping. For monoexponential mapping, radial sampling obtained a median normalized absolute deviation below 10% up to AF of 10, while Cartesian obtained this level of error only up to AF of 4. Radial sampling was also better with biexponential T_1ρ mapping, with median normalized absolute deviation below 10% up to AF of 6.

CONCLUSION

Golden-angle radial acquisitions combined with CS outperformed Cartesian acquisitions for 3D-T_1ρ mapping of knee cartilage, being it is a good alternative to Cartesian sampling for reducing scan time and/or improving image and mapping quality. The methods exponential decay models, spatio-temporal finite differences, and low rank obtained the best results for radial sampling patterns.

Whole-heart T1 mapping using a 2D fat image navigator for respiratory motion compensation

Purpose

To combine a 3D saturation‐recovery‐based myocardial T₁ mapping (3D SASHA) sequence with a 2D image navigator with fat excitation (fat‐iNAV) to allow 3D T₁ maps with 100% respiratory scan efficiency and predictable scan time.

Methods

Data from T₁ phantom and 10 subjects were acquired at 1.5T. For respiratory motion compensation, a 2D fat‐iNAV was acquired before each 3D SASHA k‐space segment to correct for 2D translational motion in a beat‐to‐beat fashion. The effect of the fat‐iNAV on the 3D SASHA T1 estimation was evaluated on the T₁ phantom. For 3 representative subjects, the proposed free‐breathing 3D SASHA with fat‐iNAV was compared to the original implementation with the diaphragmatic navigator. The 3D SASHA with fat‐iNAV was compared to the breath‐hold 2D SASHA sequence in terms of accuracy and precision.

Results

In the phantom study, the Bland‐Altman plot shows that the 2D fat‐iNAVs does not affect the T₁ quantification of the 3D SASHA acquisition (0 ± 12.5 ms). For the in vivo study, the 2D fat‐iNAV permits to estimate the respiratory motion of the heart, while allowing for 100% scan efficiency, improving the precision of the T₁ measurement compared to non‐motion‐corrected 3D SASHA. However, the image quality achieved with the proposed 3D SASHA with fat‐iNAV is lower compared to the original implementation, with reduced delineation of the myocardial borders and papillary muscles.

Conclusions

We demonstrate the feasibility to combine the 3D SASHA T₁ mapping imaging sequence with a 2D fat‐iNAV for respiratory motion compensation, allowing 100% respiratory scan efficiency and predictable scan time.

Principal component analysis fosr fast and model-free denoising of multi b-value diffusion-weighted MR images

Despite the utility of tumour characterisation using quantitative parameter maps from multi-b-value diffusion-weighted MRI (DWI), clinicians often prefer the use of the image with highest diffusion-weighting (b-value), for instance for defining regions of interest (ROIs). However, these images are typically degraded by noise, as they do not utilize the information from the full acquisition. We present a principal component analysis (PCA) approach for model-free denoising of DWI data. PCA-denoising was compared to synthetic MRI, where a diffusion model is fitted for each voxel and a denoised image at a given b-value is generated from the model fit. A quantitative comparison of systematic and random errors was performed on data simulated using several diffusion models (mono-exponential, bi-exponential, stretched-exponential and kurtosis). A qualitative visual comparison was also performed for in vivo images in six healthy volunteers and three pancreatic cancer patients. In simulations, the reduction in random errors from PCA-denoising was substantial (up to 55%) and similar to synthetic MRI (up to 53%). Model-based synthetic MRI denoising resulted in substantial (up to 29% of signal) systematic errors, whereas PCA-denoising was able to denoise without introducing systematic errors (less than 2%). In vivo, the signal-to-noise ratio (SNR) and sharpness of PCA-denoised images were superior to synthetic MRI, resulting in clearer tumour boundaries. In the presence of motion, PCA-denoising did not cause image blurring, unlike image averaging or synthetic MRI. Multi-b-value MRI can be denoised model-free with our PCA-denoising strategy that reduces noise to a level similar to synthetic MRI, but without introducing systematic errors associated with the synthetic MRI method.

Compressed sensing acceleration of biexponential 3D-T1ρ relaxation mapping of knee cartilage

PURPOSE

Use compressed sensing (CS) for 3D biexponential spin-lattice relaxation time in the rotating frame (T_1ρ ) mapping of knee cartilage, reducing the total scan time and maintaining the quality of estimated biexponential T_1ρ parameters (short and long relaxation times and corresponding fractions) comparable to fully sampled scans.

METHODS

Fully sampled 3D-T_1ρ -weighted data sets were retrospectively undersampled by factors 2-10. CS reconstruction using 12 different sparsifying transforms were compared for biexponential T_1ρ -mapping of knee cartilage, including temporal and spatial wavelets and finite differences, dictionary from principal component analysis (PCA), k-means singular value decomposition (K-SVD), exponential decay models, and also low rank and low rank plus sparse models. Synthetic phantom (N = 6) and in vivo human knee cartilage data sets (N = 7) were included in the experiments. Spatial filtering before biexponential T_1ρ parameter estimation was also tested.

RESULTS

Most CS methods performed satisfactorily for an acceleration factor (AF) of 2, with relative median normalized absolute deviation (MNAD) around 10%. Some sparsifying transforms, such as low rank with spatial finite difference (L + S SFD), spatiotemporal finite difference (STFD), and exponential dictionaries (EXP) significantly improved this performance, reaching MNAD below 15% with AF up to 10, when spatial filtering was used.

CONCLUSION

Accelerating biexponential 3D-T_1ρ mapping of knee cartilage with CS is feasible. The best results were obtained by STFD, EXP, and L + S SFD regularizers combined with spatial prefiltering. These 3 CS methods performed satisfactorily on synthetic phantom as well as in vivo knee cartilage for AFs up to 10, with median error below 15%.

3D SASHA myocardial T1 mapping with high accuracy and improved precision

Purpose

To improve the precision of a free-breathing 3D saturation-recovery-based myocardial T1 mapping sequence using a post-processing 3D denoising technique.

Methods

A T1 phantom and 15 healthy subjects were scanned on a 1.5 T MRI scanner using 3D saturation-recovery single-shot acquisition (SASHA) for myocardial T1 mapping. A 3D denoising technique was applied to the native T1-weighted images before pixel-wise T1 fitting. The denoising technique imposes edge-preserving regularity and exploits the co-occurrence of 3D spatial gradients in the native T1-weighted images by incorporating a multi-contrast Beltrami regularization. Additionally, 2D modified Look-Locker inversion recovery (MOLLI) acquisitions were performed for comparison purposes. Accuracy and precision were measured in the myocardial septum of 2D MOLLI and 3D SASHA T1 maps and then compared. Furthermore, the accuracy and precision of the proposed approach were evaluated in a standardized phantom in comparison to an inversion-recovery spin-echo sequence (IRSE).

Results

For the phantom study, Bland–Altman plots showed good agreement in terms of accuracy between IRSE and 3D SASHA, both on non-denoised and denoised T1 maps (mean difference −1.4 ± 18.9 ms and −4.4 ± 21.2 ms, respectively), while 2D MOLLI generally underestimated the T1 values (69.4 ± 48.4 ms). For the in vivo study, there was a statistical difference between the precision measured on 2D MOLLI and on non-denoised 3D SASHA T1 maps (P = 0.005), while there was no statistical difference after denoising (P = 0.95).

Conclusion

The precision of 3D SASHA myocardial T1 mapping was substantially improved using a 3D Beltrami regularization based denoising technique and was similar to that of 2D MOLLI T1 mapping, while preserving the higher accuracy and whole-heart coverage of 3D SASHA.

Electronic supplementary material

The online version of this article (10.1007/s10334-018-0703-y) contains supplementary material, which is available to authorized users.

A vectorized Levenberg-Marquardt model fitting algorithm for efficient post-processing of cardiac T1 mapping MRI

PURPOSE

T₁ mapping is an emerging MRI research tool to assess diseased myocardial tissue. Recent research has been focusing on the image acquisition protocol and motion correction, yet little attention has been paid to the curve fitting algorithm.

METHODS

After nonrigid registration of the image series, a vectorized Levenberg-Marquardt (LM) technique is proposed to improve the robustness of the curve fitting algorithm by allowing spatial regularization of the parametric maps. In addition, a region-based initialization is proposed to improve the initial guess of the T₁ value. The algorithm was validated with cardiac T₁ mapping data from 16 volunteers acquired with saturation-recovery (SR) and inversion-recovery (IR) techniques at 3T, both pre- and post-injection of a contrast agent. Signal models of T₁ relaxation with 2 and 3 parameters were tested.

RESULTS

The vectorized LM fitting showed good agreement with its pixel-wise version but allowed reduced calculation time (60 s against 696 s on average in Matlab with 256 × 256 × 8(11) images). Increasing the spatial regularization parameter led to noise reduction and improved precision of T₁ values in SR sequences. The region-based initialization was particularly useful in IR data to reduce the variability of the blood T₁.

CONCLUSIONS

We have proposed a vectorized curve fitting algorithm allowing spatial regularization, which could improve the robustness of the curve fitting, especially for myocardial T₁ mapping with SR sequences.

Accelerating 3D-T1ρ mapping of cartilage using compressed sensing with different sparse and low rank models

PURPOSE

To evaluate the feasibility of using compressed sensing (CS) to accelerate 3D-T_1ρ mapping of cartilage and to reduce total scan times without degrading the estimation of T_1ρ relaxation times.

METHODS

Fully sampled 3D-T_1ρ datasets were retrospectively undersampled by factors 2-10. CS reconstruction using 12 different sparsifying transforms were compared, including finite differences, temporal and spatial wavelets, learned transforms using principal component analysis (PCA) and K-means singular value decomposition (K-SVD), explicit exponential models, low rank and low rank plus sparse models. Spatial filtering prior to T_1ρ parameter estimation was also tested. Synthetic phantom (n = 6) and in vivo human knee cartilage datasets (n = 7) were included.

RESULTS

Most CS methods performed satisfactorily for an acceleration factor (AF) of 2, with relative T_1ρ error lower than 4.5%. Some sparsifying transforms, such as spatiotemporal finite difference (STFD), exponential dictionaries (EXP) and low rank combined with spatial finite difference (L+S SFD) significantly improved this performance, reaching average relative T_1ρ error below 6.5% on T_1ρ relaxation times with AF up to 10, when spatial filtering was used before T_1ρ fitting, at the expense of smoothing the T_1ρ maps. The STFD achieved 5.1% error at AF = 10 with spatial filtering prior to T_1ρ fitting.

CONCLUSION

Accelerating 3D-T_1ρ mapping of cartilage with CS is feasible up to AF of 10 when using STFD, EXP or L+S SFD regularizers. These three best CS methods performed satisfactorily on synthetic phantom and in vivo knee cartilage for AFs up to 10, with T_1ρ error of 6.5%.