3D SASHA myocardial T1 mapping with high accuracy and improved precision
To improve the precision of a free-breathing 3D saturation-recovery-based myocardial T1 mapping sequence using a post-processing 3D denoising technique.
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).
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).
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.
KeywordsMyocardial T1 mapping Precision Accuracy Denoising Cardiac MRI
Quantitative myocardial T1 mapping has been explored as a tool to evaluate different cardiomyopathies, particularly to assess myocardial fibrosis .
Several T1 mapping imaging techniques have been proposed which use inversion recovery [2, 3], saturation recovery [4, 5] or a combination of both pre-pulses . The majority of these sequences are employed as two-dimensional (2D) imaging techniques where a single 2D map is acquired per breath-hold. However, breath-holding limits the achievable spatial resolution or may introduce slice mis-registration artifacts, which can hinder fibrosis detection in the right ventricle, atrial wall or apex of the left ventricle. A three-dimensional (3D) T1 mapping technique would thus be preferable, as they permit volumetric coverage of the heart in free-breathing with higher signal-to-noise ratio (SNR) and improved image resolution.
Recently, a 3D saturation-recovery-based T1 mapping imaging technique, called 3D SASHA, has been proposed, which permits the acquisition of whole-heart images under free-breathing conditions . This technique has been successfully validated in healthy subjects and showed higher accuracy in the estimation of myocardial T1 than the 2D modified Look-Locker inversion recovery-based (MOLLI) technique used clinically. Although the precision of the 3D SASHA T1 map has been shown to be superior to 2D SASHA, it is still lower than that of 2D MOLLI. This lower precision is due in part to the low dynamic range of signal intensity when using saturation pulses, which makes saturation-recovery-based T1 mapping sequences more sensitive to noise. A precise estimation of myocardial T1 values is important for achieving higher reproducibility and ultimately improving diagnostic utility.
To address this issue, efficient 2D denoising techniques have recently been proposed to reduce noise in the native T1-weighted images while preserving the inherent cardiac structures [8, 9]. Jang et al.  employed 2D denoising based on a tissue similarity distance to reduce the sensitivity to noise of 2D STONE  T1 mapping when three-parameter fitting is employed. Bustin et al.  developed a 2D denoising method using the Beltrami regularization to improve the precision of 2D breath-hold saturation-based myocardial T1 mapping sequences while maintaining accuracy of T1 quantification. This feature-preserving regularization technique can greatly reduce the staircase effect  often associated with other regularity priors. In this study, we investigated the use of the Beltrami regularization denoising approach to further improve the precision and image quality of the 3D free-breathing SASHA sequence. The Beltrami denoising framework will be investigated to exploit the spatio-contrast variation correlation in the native 3D T1-weighted images.
Materials and methods
The Beltrami regularization for image denoising and enhancement was introduced for 2D natural images by Sochen et al.  and was proposed for 2D MRI myocardial T1 mapping denoising by Bustin et al. . The potential of the Beltrami regularization framework lies in the general definition of the space-feature manifold and the choice of its metric. In particular, the regularization can be chosen such that the energy of the unknown image corresponds to an arbitrary interpolation between quadratic or total variation gradient penalties.
In this work, we apply the Beltrami denoising framework to 3D T1 mapping by exploiting the 3D spatial and T1 recovery dimension redundant information. This approach thus exploits the 3D spatial gradients of each T1-weighted image and the common edge information between T1-weighted images with varying contrast. A coupling between the T1-weighted images is performed as described by Bustin et al. , to enforce common edge information across the 3D images in the T1 encoding direction. With this approach, common structures/edges in different T1-weighted images are preserved, while local intensity variations specific to each T1-weighted image are treated as noise, and thus reduced. Therefore, edges/details in the images will be better enhanced if they are present in multiple T1-weighted images.
The optimization problem described above was solved using an efficient and simple primal–dual hybrid gradient algorithm [12, 13]. The parameters \(\lambda\) and \(h\) were set to 0.25 and 5 respectively, for all the experiments. These parameters were empirically optimized on three representative datasets to provide the best image quality and then the same parameters were employed in the remaining cases. A weight of 1.0 was employed for the Beltrami constant as indicated in Zosso and Bustin .
All imaging studies were performed on a 1.5 T MR scanner (Ingenia, Philips, Best, The Netherlands). T1 mapping was performed using the saturation-based 3D SASHA sequence described by Nordio et al. . Phantom and cardiac studies used a 32-channel phased-array coil. The images from phantom and healthy subjects were collected from the scanner in DICOM format and were loaded into a computer for offline image processing and analysis using MATLAB (MathWorks, Natick, MA, USA).
To evaluate the performance of the proposed 3D denoising algorithm with regard to the accuracy and precision of T1 values, a standardized phantom containing nine agar/NiCl2 vials was used, with T1 values ranging from 250 to 1500 ms . The phantom was imaged using the gold-standard 2D inversion-recovery spin echo (IRSE), the 2D MOLLI (3-(3)-3-(3)-5) and the proposed 3D SASHA imaging techniques. Acquisitions parameters for the 2D IRSE sequence were: field of view (FOV) = 150 × 150 mm2, image resolution = 1.95 × 1.95 mm2, 10 mm slice thickness, 20 inversion times (TI) varying from 50 to 5000 ms, TR(i) = 7000 ms plus the variable inversion time TI(i), and TE = 6 ms. Acquisition parameters for the 2D MOLLI sequence included: FOV = 300 × 280 mm, image resolution = 1.7 mm × 2.1 mm, slice thickness = 10 mm, TR/TE = 2.6/1.3 ms, FA = 35°, acquired matrix = 176 × 132, scan time of 12 s. The acquisition parameters for the 3D SASHA sequence were: FOV = 300 × 300 × 90 mm, TR/TE = 3.2/1.6 ms, image resolution = 1.4 mm × 1.4 mm × 8 mm, FA = 35°, acquired matrix = 216 × 214, saturation time within the range of 100–700 ms for a heart rate of 60 beats per minute (bpm), parallel imaging with a SENSE factor of 2 in the phase-encoding direction, nominal scan time approximately 4 min. Nine T1-weighted images were acquired, including one image without any magnetization preparation and eight images at different saturation times. All three imaging techniques were reconstructed to the same in-plane resolution of 1.25 × 1.25 mm2.
In vivo experiments
All subjects recruited to this study provided written informed consent, with study approval from the institutional review board (1/11/12). Fifteen healthy subjects with no history of cardiovascular disease underwent imaging using the 2D MOLLI and 3D SASHA sequences in the short-axis plane, using the same imaging parameters as for the phantom experiment. A total of 11 slices were acquired for the 3D SASHA sequence, with an oversampling factor of 1.14. The acquisition of the 3D SASHA sequence was performed in free-breathing with a nominal scan duration of 4:14 (min:s) for a heart rate of 60 bpm and 100% respiratory scan efficiency. A 1D diaphragmatic navigator was used for respiratory motion compensation, with a gating window of 5 mm and a tracking factor of 0.6 in the foot–head direction.
The 3D Beltrami denoising was applied to the magnitude 3D T1-weighted images of the 3D SASHA acquisition prior to performing the T1 fitting. T1 maps were reconstructed offline using MATLAB with a three-parameter fitting model for the T1-weighted images before and after denoising . A spatially matching region of interest (ROI) was manually drawn on the T1 map of the 2D IRSE (phantom only), 2D MOLLI, and 3D SASHA before (“non-denoised 3D SASHA”) and after (“denoised 3D SASHA”) denoising in all vials for the phantom experiment as well as in the septum of the myocardium for the in vivo data. The mean and standard deviation of the T1 measurements were calculated and used as an estimation of the accuracy and precision of the respective T1 mapping techniques. A Bland–Altman plot was used to compare the different imaging techniques for the phantom study. The precision measured on the 2D MOLLI and non-denoised and denoised 3D SASHA T1 maps of the healthy subjects were compared using a Mann–Whitney U test. An American Heart Association (AHA) segmentation  was used to compare the accuracy and precision of the 3D SASHA before and after denoising and for a 3D visualization of the left ventricle. The cardiac volume was represented in 16 segments and 3 slices (apex, mid and base), with the segment 17 corresponding to the blood pool. The value in each segment was averaged within the fifteen healthy subjects.
For statistical analysis, GraphPad Prism v5 for Windows (GraphPad Software, La Jolla, CA, USA) was used. Two-tailed values of P < 0.05 were considered statistically significant.
In vivo experiments
The total scan time for the 3D SASHA acquisition in healthy subjects was of 12:4 ± 1:1 (min:s), with an average scan efficiency of 36% and heart rates of 60–70 bpm.
The impact of the choice of the parameters h and λ on the image quality of the denoised 3D SASHA T1 map is shown in the supplementary Figure S4. The denoised T1 map becomes sharper with higher values for the parameter h. However, a value of h that is too high may result in patchy artifacts. The parameters \(\lambda\) and \(h\) were set to 0.25 and 5, respectively, for all the experiments.
The processing time required for the Beltrami denoising on one 3D cardiac volume was about 1 min.
In this study, we introduced a 3D free-breathing SASHA T1 mapping framework with high accuracy and improved precision following implementation of a denoising step. 3D SASHA T1 mapping is characterized by higher spatial resolution and volumetric coverage and higher accuracy than the 2D MOLLI technique used clinically; however, its precision is lower [4, 7]. Here we have shown that it is feasible to obtain precision comparable to that of the 2D MOLLI technique while maintaining high accuracy and whole-heart coverage characteristic for 3D SASHA. This is achieved by applying a Beltrami denoising method to the T1-weighted 3D SASHA images prior to the T1 fitting. The denoising technique, proposed for 2D T1 mapping in Bustin et al. , imposes edge-preserving regularity and exploits the co-occurrence of 3D spatial gradients in the native images by incorporating a multi-contrast Beltrami regularization.
In contrast to the technique presented in Weingartner et al. , where multi-slice denoising using local neighborhood averaging for each pixel was performed, we utilized 3D spatial gradients to exploit the redundancy and sparsity of the whole 3D volume and Beltrami regularization to exploit sparsity in the T1 recovery dimension. This sparse formulation has the advantage of being robust to high levels of noise and maintaining sharp edges, while avoiding computationally expensive reconstruction often associated with local filters. Compared to the 2D denoising technique introduced in Bustin et al. , our framework further exploits redundancy and sparsity in the whole 3D volume and provides whole-heart volumetric coverage with high in-plane spatial resolution.
As demonstrated in the phantom experiment, the accuracy of the 3D SASHA T1 map before and after denoising is comparable to the gold-standard IRSE technique, with a slight increase of a few milliseconds for some vials. In terms of precision, the improvement is small, due to the small range of standard deviation. Conversely, with the 2D MOLLI imaging technique, the myocardial T1 was underestimated, as shown previously [2, 17]. In vivo experiments in healthy subjects demonstrated preserved accuracy of non-denoised and denoised 3D SASHA. In addition, the precision was considerably improved after denoising, with no statistical difference compared to 2D MOLLI T1 maps. The AHA segmentation allows for better visualization of the effect of the 3D denoising method on the whole volume, in terms of both accuracy and precision. The precision in general is improved across the whole left ventricle, as well as in the blood pool indicated in the central segment. In fact, the denoising should have a larger effect in areas with more noise or noise-like structures. As the blood pool signal is more inhomogeneous, we expect the denoising to be more efficient in that area. The variability in the precision measured in the myocardial inferolateral wall could be explained by the susceptibility artifacts and some residual motion . After denoising, the delineation of the myocardial borders and papillary muscles is maintained, while the signal in general is more homogeneous. For respiratory motion compensation, the 3D SASHA imaging sequence is combined with a 1D diaphragmatic navigator gating and tracking. The navigator generally allows for respiratory scan efficiency of ~ 50%; however, markedly lower scan efficiency (20–30%) is often observed in patients with more irregular breathing patterns. An alternative respiratory motion compensation technique, such as image-based navigation or self-navigation [19, 20], would permit 100% respiratory scan efficiency, thus reducing the scan time and consequently potential additional bulk motion artifacts associated with lengthy scans. Compared to 3D SASHA, 2D MOLLI permits the acquisition of one slice in only 12 s of breath-hold. However if several slices are acquired, resting periods of at least the same duration as the breath-hold are necessary between the acquisition of each slice. Consequently, to acquire several slices and to achieve the same volumetric coverage of the 3D SASHA presented here, 2D MOLLI would require a total scan time of about 4 min. In addition, 2D MOLLI has lower image resolution than 3D SASHA; a higher resolution would require a longer acquisition window, making the MOLLI scan more susceptible to cardiac motion.
The denoising approach applied in this study is a post-processing step and thus comes at no additional acquisition or reconstruction cost. The additional post-processing time of ~ 1 min per 3D volume to execute the denoising technique is reasonable and, if clinically validated, inline implementation would be feasible. The integration of the denoising algorithm in the reconstruction step may accelerate and simplify the whole framework and will be investigated in future work. Overall, 3D SASHA with denoising appears to be a promising alternative for providing accurate T1 values while reducing variability, thus addressing current barriers to the wide-scale adoption of quantitative myocardial tissue characterization techniques in the clinical setting.
We have demonstrated the feasibility of the 3D Beltrami denoising technique to preserve accuracy and improve precision in myocardial 3D SASHA T1 mapping. Ultimately, this technique could enable accurate and more precise reconstruction of myocardial T1 maps, with the potential to offer better visual image analysis and improved performance of post-processing procedures such as image registration and segmentation.
This work was supported by (1) the King’s College London and Imperial College London EPSRC Centre for Doctoral Training in Medical Imaging (EP/L015226/1) and (2) EPSRC grants EP/P001009/1 and EP/P007619 and (3) FONDECYT no. 1161051, (4) the Wellcome EPSRC Centre for Medical Engineering (NS/A000049/1 and WT 203148/Z/16/Z), and (5) the Department of Health via the National Institute for Health Research (NIHR) Cardiovascular Health Technology Cooperative (HTC) and comprehensive Biomedical Research Centre awarded to Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London and King’s College Hospital NHS Foundation Trust. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.
GN: study conception and design, acquisition of the data, analysis and interpretation of data, drafting of the manuscript. AB: study conception and design, analysis and interpretation of data, drafting of the manuscript. RMB: study conception and design, analysis and interpretation of data, critical revision. CP: study conception and design, analysis and interpretation of data, critical revision. MH: analysis and interpretation of data, critical revision. AC: critical revision. TFI: critical revision. IR: critical revision. FO: study conception and design, critical revision.
Compliance with ethical standards
Conflict of interest
All the other authors declare that they do not have any conflict of interests.
The study was approved by the National Research Ethics Service and written informed consent was obtained for all the subjects.
- 3.Piechnik SK, Ferreira VM, Dall’Armellina E, Cochlin LE, Greiser A, Neubauer S, Robson MD (2010) Shortened modified Look-Locker inversion recovery (ShMOLLI) for clinical myocardial T1-mapping at 1.5 and 3 T within a 9 heartbeat breathhold. J Cardiovasc Magn Reson 12:69. https://doi.org/10.1186/1532-429X-12-69 CrossRefPubMedPubMedCentralGoogle Scholar
- 6.Weing S, Akc M, Basha T, Kissinger KV, Goddu B, Berg S, Manning WJ, Nezafat R (2014) Combined saturation/inversion recovery sequences for improved evaluation of scar and diffuse fibrosis in patients with arrhythmia or heart rate variability. Magn Reson Med 1034:1024–1034. https://doi.org/10.1002/mrm.24761 CrossRefGoogle Scholar
- 9.Jang J, Nakamori S, Nezafat R (2017) Improving precision of myocardial T1 mapping with 3-parameter model using tissue characteristic-based denoising. In: Proceedings of the 25th annual meeting of ISMRM Hawaii USAGoogle Scholar
- 11.Zosso D, Bustin A (2014) A primal-dual projected gradient algorithm for efficient Beltrami regularization. UCLA CAM rep, pp 14–52Google Scholar
- 13.Zhu M, Chan T (2008) An efficient primal-dual hybrid gradient algorithm for total variation image restoration. Technical report, UCLA CAM report 08-34Google Scholar
- 16.Cerqueira MD, Weissman NJ, Dilsizian V, Jacobs AK, Kaul S, Laskey WK, Pennell DJ, Rumberger JA, Ryan T, Verani MS (2002) Standardized myocardial segmentation and nomenclature for tomographic imaging of the heart. Circ Cardiovasc Imaging 105(4):539–542Google Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.