Synopsis

Keywords: Myocardium, Data Processing

Cine cardiac magnetic resonance (CMR) imaging is an optimal cardiac volumetric analysis method because of its high contrast resolution. However, its spatial resolution is limited owing to prolonged scanning and breath-holding. Although compressed sensing–sensitivity encoding (Compressed SENSE; CS) and its deep-learning-based advancement (SmartSpeed AI; SSAI) can reduce the scan time, the spatial resolutions remain unchanged. Herein, we investigated the effect of a deep-learning-based super-resolution technique (SmartSpeed Precise Image; SSPI) on the cine CMR visual image quality in comparison with CS and SSAI with conventional zero-filling interpolation; resultantly, the SSPI significantly improved the visual image quality scores.

Introduction

Cine cardiac magnetic resonance (CMR) based on a balanced steady-state free precession pulse sequence (bSSFP) is a widely used and essential sequence for CMR. Wall motion abnormalities, ventricular volumes, and ejection fractions of the left and right ventricles (LV and RV, respectively) can be evaluated using cine CMR¹. Although cine CMR is the gold standard for volumetric analysis because of its high contrast resolution, its spatial resolution is lower than that of ultrasonography. Prolonged scanning and breath-holding times make it difficult to improve the spatial resolutions of two-dimensional bSSFP. Although compressed sensing–sensitivity encoding (Compressed SENSE; CS) is applied to accelerate image acquisition, a higher acceleration factor results in poorer image quality. A deep-learning (DL) (Adaptive-CS-Net)-based image reconstruction technique (SmartSpeed AI; SSAI) has the potential to improve the image quality². However, because SSAI cannot affect spatial resolution, its impact on image quality is limited. The conventional zero-filling interpolation (ZIP) technique is widely used to improve “apparent” spatial resolution; however, it cannot improve the actual spatial resolution. Recently, a DL-based super-resolution technique (SSPI) has been developed to improve actual spatial resolution. We hypothesized that this DL-based super-resolution technique would reduce the scan time while improving the image quality. This study aimed to compare the image quality of cine CMR images reconstructed using CS with ZIP (CS-ZIP), SSAI with ZIP (SSAI-ZIP), and SSAI with SSPI (SSAI-SSPI).

Methods

This prospective study was approved by the institutional review board. Written informed consent was obtained from all the participants. Six healthy volunteers (all men; median age, 36 years (range, 32–57 years)) were scanned using a 3.0T-MRI scanner (Ingenia Elition X, Philips Healthcare, Best, the Netherlands) with a dS Torso coil. The bSSFP images with the left ventricular short-axis plane were acquired with two different acceleration factors (acceleration factors of 2 and 4) and reconstructed using three different reconstruction methods (CS-ZIP, SSAI-ZIP, and SSAI-SSPI). Thus, six types of images were reconstructed. The acquisition parameters were as follows: TR/TE (ms) = 3.8/1.79, Flip angle (degree) = 70, Slice thickness (mm) = 10, Field of view (mm) = 300×300, acquisition matrix (mm) = 2.0×2.0, reconstruction matrix (mm) = 0.94×0.94, number of phases per cardiac cycle (phases) = 20.
SSAI is a convolutional neural network called Adaptive-CS-Net² that allows the reconstruction of images acquired with CS-based variable density undersampling patterns³. SSPI is a DL-based reconstruction technique combined with Adaptive-CS-Net and Precise Image Net, which is a DL-model applied to remove ringing artifacts and to replace the traditional zero-filling strategy to increase the matrix size and thus the image sharpness; these types of networks are known as SuperResolution networks^4,5.
The acquisition time per slice for each acquisition method was measured. The visual image quality scores using whole cardiac phase images were assessed by two independent radiologists (seven and two years of experience in cardiovascular imaging) using a 3-point scale score (3=excellent, 2=moderate, 1=poor) with each of four categories: the sharpness between the LV or RV free wall and the ventricular cavity, the sharpness between LV or RV trabeculae and the ventricular cavity, the overall image quality considering image noise, and any artifact. All variables were presented as medians (ranges). The summed scores of the two observers were compared using Wilcoxon’s signed-rank test with Bonferroni correction because of the multiple comparison test problems. Inter-observer reproducibility was assessed using kappa-statistics (0.0–0.20, poor; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, good; 0.81–1.0, excellent). A p-value < 0.00033 was considered statistically significant based on a fifteen-times statistical test from six types of images.

Results

The heart rate was 60 (52–65) per minute. The acquisition time was 6.0 (5.3–6.3) seconds with an acceleration factor of 2 and 3.9 (3.3–4.0) seconds with an acceleration factor of 4. SSPI significantly improved the visual image quality scores regarding the sharpness between the myocardial wall and cavity (Fig. 1), sharpness between the trabeculae and cavity (Fig. 2), and overall quality (Fig. 3) in comparison with conventional ZIP. A representative case is shown in Fig. 4. The scores for any artifacts were not significant among the six types of images. Although inter-observer reproducibility of the scores of the overall quality was poor (kappa=0.18), the others were moderate or good (kappa = 0.54 and 0.58 for the sharpness of the LV and RV myocardia, respectively; 0.70 and 0.50 for the sharpness of the LV and RV trabeculae, respectively; 0.48 for artifacts).

Discussion

The novel DL-based reconstruction method, SSPI, provides better image quality than that of the conventional ZIP. Moreover, the image quality of a doubled acceleration factor with SSPI is better than that of the initial acceleration factor with ZIP. Furthermore, SSPI neither removed nor worsened the artifacts. SSPI with a higher acceleration factor may improve the image quality of cine CMR, while reducing the imaging time.

Conclusion

The novel DL-based super-resolution method (SSPI) could improve the image quality of cine CMR in comparison with conventional ZIP.

Acknowledgements

None.