Table 1.
Summary of reconstruction techniques for 2D cine CMRI
| Authors | Year | Mode | Method | Salient features | Performance |
|---|---|---|---|---|---|
| Tsao et al. [59] | 2003 | Multi-BH | k-t BLAST, k-t SENSE | Model-based method exploiting data correlations to recover unacquired samples. Cartesian sampling | 4-fold acceleration. Spatial res., 2.42 × 2.52 mm2 (slice thickness 10 mm). Temporal res., 26 ms |
| Huang et al. [60] | 2005 | Multi-BH | k-t GRAPPA | GRAPPA combined with sliding window techniques for missing data interpolation. Cartesian sampling | AF = 7. Reduction factor, 5.17. Spatial res., 1.77 × 1.82 mm2 (slice thickness, 6 mm). Number of phases, 14. Reconstruction time, 4 s per frame |
| Lustig et al. [61] | 2006 | Multi-BH | k-t SPARSE | CS-based method exploiting spatial and temporal sparsity of data. Cartesian sampling | 7-fold frame-rate acceleration. Spatial res., 2.5 × 2.5 mm2 (slice thickness, 9 mm). Temporal res., 40 ms. Reconstruction time, 1 h per 64 × 64 × 64 scene |
| Jung et al. [62] | 2009 | Multi-BH | k-t FOCUSS | CS method with ME-MC based on block matching. Cartesian sampling | AF = 6. Spatial res., 1.25 × 1.17 mm2 (slice thickness, 5 mm). 25 cardiac phases |
| Pedersen et al. [63] | 2009 | Multi-BH | k-t PCA | Generalization of k-t BLAST/SENSE using PCA temporal constraint. Cartesian sampling | Myocardial perfusion images acquired in a pig. 8-fold acceleration. Spatial res., 1.25 × 1.25 mm2 (slice thickness, 10 mm). 64 frames |
| Christodoulou et al. [64] | 2010 | Multi-BH | PSF | Partially separable function reconstruction with anatomical constraints | Data of rat hearts. Spatial res., 390 μm in-plane (slice thickness, 1.5 mm). Temporal res., 15 ms |
| Lingala et al. [65] | 2011 | Multi-BH | k-t SLR | Low-rank structure using KLT to exploit the sparsity. Cartesian sampling | Cardiac perfusion MRI data. AF = 11. Matrix size, 90 × 190 |
| Hansen et al. [66] | 2012 | FB real-time | – | Temporal multi-resolution scheme combining PI with MC based on nonrigid registration. Cartesian and golden-angle radial sampling | 2-fold PI acceleration. Spatial res., 1.4–1.5 × 1.9–2 mm2 (Cartesian), 1.4–1.5 × 1.4–1.5 mm2 (golden angle radial), slice thickness, 6 mm. Temporal res., 30 ms |
| Asif et al. [67] | 2013 | Multi-BH | MASTER | CS with ME-MC based on motion-adaptive spatio-temporal regularization. Cartesian sampling | Retrospective downsampling with reduction factor up to 10. Spatial res., 1.56 × 1.37 mm2 (slice thickness, 12 mm). 16 cardiac phases |
| Feng et al. [68] | 2013 | FB real-time |
k-t SPARSE SENSE |
Combination of CS and PI for real-time imaging. Cartesian sampling | 8-fold acceleration. Spatial res., 2.3 × 2.3 mm2 (slice thickness, 8 mm). Temporal res., 43.2 ms. Offline reconstruction time, 4.6 min per slice |
| Usman et al. [69] | 2013 | FB | MC-CS | Generalized MC in CS reconstruction. Respiratory motion self-gating by low resolution virtual 2D navigator images. Golden angle radial sampling | AF = 4–6. Spatial res., 1.5–2 × 1.5–2 mm2. 20 cardiac phases. Temporal res., 30–40 ms. Reconstruction time, 2–2.5 h. |
| Xue et al. [70] | 2013 | FB real-time | – | SPIRiT non-linear reconstruction with spatial-temporal regularization (Harr wavelet transformation) and ME-MC based on non-rigid registration. Cartesian time-interleaved sampling | Scan time, 16–20 s per acquired slice. PI reduction factor of R = 4. Spatial res., 1.3–1.8 × 1.8–2.1 mm2 (slice thickness, 8 mm). 30 cardiac phases. Temporal res., 34.3 ± 9.1 ms. Inline reconstruction time (Gadgetron), 80–120 s per slice |
| Schmidt et al. [71] | 2013 | FB real-time | rtCS11 | Real-time CS-based reconstruction with k-t regularization. Cartesian sampling | Scan time, 1 heartbeat. AF = 10.9. Spatial res., 1.7 × 1.7 mm2 (slice thickness, 6 mm). Temporal res., 30 ms. Online reconstruction |
| Trémoulhéac et al. [72] | 2014 | Multi-BH | k-t RPCA | L+S decomposition based on RPCA with temporal FT. Variable density Cartesian and pseudo-radial sampling | AF = 8. Matrix size, 128 × 128 (90 frames). Reconstruction time, 10 min |
| Wang et al. [73] | 2014 | Multi-BH | – | CS-based reconstruction with DL. Retrospective Cartesian undersampling | AF up to 8. Matrix size, 150–256 × 256–304 (14–26 frames). Reconstruction time, 11.3–24.3 min |
| Vincenti et al. [74] | 2014 | Single-BH | – | CS-based method with Cartesian acquisition | AF = 11. 3 long-axis and 4 short-axis views. Spatial res., 1.5 × 1.5 mm2 (slice thickness, 6 mm). 24 cardiac phases. Temporal res., 30 ms. BH duration, 14 s |
| Royuela-del Val et al. [75] | 2015 | Single-BH | kt-WiSE | MC-CS based on GW registration with SENSE. Golden angle radial sampling | AF = 16. Spatial res., 2 × 2 mm2, (slice thickness, 8 mm, 12 slices). 16 cardiac phases Temporal res., 46.4 ms. BH duration, 11.1 s |
| Velikina et al. [76] | 2015 | Multi-BH | MOCCO | Pre-estimated low-rank temporal signal models. Variable density Cartesian sampling | AF up to 15. Spatial res., 1 × 1.7 mm2 (26 and 30 cardiac phases) |
| Otazo et al. [44] | 2015 | Multi-BH | – | L+S reconstruction. Cartesian sampling for cardiac cine. Radial sampling for abdominal and breast DCE-MRI | 8-fold acceleration. Spatial res., 1.25 × 1.25 mm2 (slice thickness, 8 mm). 24 temporal frames |
| Poddar and Jacob [77] | 2016 | FB real-time | SToRM | Manifold smoothness regularized reconstruction with radial sampling | Scan time, 42 s per slice. Spatial res., 1.17 × 1.17 mm2, (slice thickness, 5 mm, 5 slices). Temporal res., 42 ms. Reconstruction time, 24 min (l2-SToRM) and 4.9 h (l1-SToRM) |
| Royuela-del Val et al. [78] | 2016 | Multi-BH | GW-CS | CS method with ME-MC based on non-rigid GW registration and Cartesian sampling | AF up to 12. Spatial res., 2 × 2 mm2 (slice thickness, 8 mm). 16 cardiac phases |
| Miao et al. [79] | 2016 | Single-BH | LLR + FD | Locally low rank with temporal finite difference and PI using golden-angle radial sampling | AF = 19–23. Spatial res., 2 × 2 mm2, (slice thickness, 8 mm, 12 SA slices). Temporal res., 40 ms (19–20 time frames). BH duration, 9–13 s |
| Feng et al. [80] | 2016 | FB | XD-GRASP | CS-based reconstruction of extra cardio-respiratory motion states. Continuous acquisition with golden-angle trajectory | Scan time, 20 s per slice. AF = 16. Spatial res., 2 × 2 mm2, (slice thickness, 8 mm, 3 SA + 1 4CH slices). Temporal res., 45 ms. 18–26 cardiac phases and 10–16 respiratory phases |
| Chen et al. [81] | 2016 | FB real-time | – | Parallel online reconstruction using dTV and accelerated reweighted least squares algorithm. Radial sampling | Matrix size, 256 × 256 × 24. Reconstruction time, 33.1 s |
| Royuela-del Val et al. [82] | 2017 | Multi-BH | JW-tTV | CS-MC method using Jacobian weighted temporal TV as sparse regularization term. Cartesian sampling | AF = 12. FOV = 320 × 320 mm2, (slice thickness, 8 mm). 30 cardiac phases |
| Mohsin et al. [83] | 2017 | Multi-BH | PRICE | Implicit inter-frame MC based on patch smoothness regularization. Cartesian sampling | Scan time, two heartbeats per slice. AF = 6. Spatial res., 2.5 × 2.5 mm2. 1 slice, 20 temporal frames (16 lines per frame). Reconstruction time, 7 min |
| Chen et al. [84] | 2017 | Multi-BH | – | L+S method. RPCA inverse problem solved by IALM. Cartesian and pseudo-radial sampling | AF = 6. Spatial res., 1.25 × 1.25 mm2, (slice thickness, 10 mm). 1 slice, 30 temporal frames. Reconstruction time, 2–2.2 min |
| Xu et al. [85] | 2017 | Multi-BH | G-NADM, L-NADM | L+S method with NADM for nonconvex RPCA | Matrix size, 256 × 256. 1 slice, 24 temporal frames. Reconstruction time, 3–3.3 min |
| Roohi et al. [86] | 2017 | Multi-BH | k-t MLSD | Multi-dimensional L+S decomposition method. Cartesian and radial sampling | Sampling rate, 0.25. Spatial res., 1.35 × 1.05 mm2, (slice thickness, 10 mm). 25 temporal frames (66 bpm). Reconstruction time, 26.64 s per slice |
| Royuela-del Val et al. [87] | 2017 | Single-BH | JW-tTV-GR | Adaptation of JW-tTV to golden radial acquisition pattern. Whole-heart coverage | AF = 16. 12–14 SA slices. Spatial res., 2 × 2 mm2 (slice thickness, 8 mm). 13–16 cardiac phases. Temporal res., 46.4 ms. BH duration, 10–13 s |
| Wang et al. [88] | 2017 | FB real-time | PDLDTV | Parallel DicL and dTV method using a primal-dual algorithm. Radial sampling | Sampling rate, 70% 1st frame, 15% rest. Matrix size, 256 × 256, 24 temporal frames. Reconstruction time, 2 min |
| Tolouee et al. [89] | 2018 | Multi-BH | – | L+S method with MC based on a deformable registration method. Cartesian sampling | AF = 12. Spatial res., 1.35 × 1.05 mm2, (slice thickness, 10 mm). Temporal res., 25 ms |
| Li et al. [90] | 2018 | FB real-time | – | k-space variant reduced-FOV reconstruction. Radial sampling | Spatial res., 1.7 mm2, (slice thickness, 8 mm). Temporal res., 40 ms. Reconstruction time, 2 s per frame |