4894

Reconstructing real-time exercise stress cine images with multiple sets of sensitivity maps

Chong Chen¹, Yingmin Liu¹, Orlando Simonetti¹, and Rizwan Ahmad¹

¹The Ohio State University, Columbus, OH, United States

Synopsis

Motivated by the changing sensitivity maps due to exaggerated chest wall motion in real-time stress cardiac MRI (CMR), we utilize ESPIRiT-based multiple sets of sensitivity maps in a SENSE-based reconstruction method to reduce image artifact. The proposed method was tested on twelve volunteers and compared with the images reconstructed using a temporally invariant single set of sensitivity maps as well as a temporally varying single set of sensitivity maps. It was demonstrated that using multiple sets of sensitivity maps leads to significant reduction in image artifact.

Background

Some cardiac conditions that do not present at rest may be identified via cardiac MRI stress testing (stress CMR). Cine imaging can be used to assess cardiac function in stress CMR; however, patients typically cannot hold their breath post-exercise, and therefore segmented cine is not a viable option. Traditional GRAPPA-based real-time imaging has relatively poor spatial resolution (~3.5mm) and temporal resolution (~65ms) that may be inadequate at post-stress heart rates. To achieve acceptable spatial and temporal resolution, compressive sensing (CS) inspired methods have been proposed for stress CMR; however, these methods do not address the fundamental issue that stress CMR data, due to exaggerated diaphragm and chest wall motion, cannot be accurately represented by a single set of sensitivity maps or GRAPPA kernels. Noting that sensitivity maps will exhibit temporal variability due to respiratory motion, we use real-time cine data from twelve volunteers to demonstrate that employing ESPIRiT-based multiple sets of sensitivity maps [1] can significantly reduce artifacts in stress CMR.

Method

A nine-slice stack of short axis real-time cine images covering the left ventricle was acquired in twelve healthy volunteers immediately following maximal exercise to exhaustion on an MRI compatible treadmill positioned next to the MRI patient table. The protocol information on a 1.5T scanner (Avanto, Siemens Healthcare, Erlangen, Germany) was: FOV 350-420mm, TE/TR 1.02/2.34ms, image matrix 192x126, flip angle 60 degrees, acceleration rate 9 using VISTA sampling pattern [2], spatial resolution ~2.5x2.2mm², temporal resolution 33ms.

Images were reconstructed using a SENSE-based CS method with auto-tuning of regularization parameters [3]. Cardiac frames were averaged to generate ACS lines, and ESPIRiT was used to estimate the coil sensitivity maps. When using multiple sets of sensitivity maps, the strict SENSE model was extended to “soft” SENSE reconstruction, where the signal in k-space was represented in terms of multiple sets of images, each with its own set of sensitivity maps. The final reconstructed image was the sensitivity map-weighted sum of all sets of images [1].

As a comparison, we reconstructed the images using three different sensitivity maps: a single set of sensitivity maps (SS), a single set of temporally varying (respiratory phase specific) sensitivity maps (TSS), and temporally invariant multiple sets of sensitivity maps (MS). One and three sets of sensitivity maps were estimated for SS and MS, respectively, using the ACS lines generated by averaging undersampled k-space data from all cardiac frames. For TSS, one set of sensitivity maps was estimated from ACS lines generated by averaging undersampled k-space data from 17 consecutive frames, yielding temporally varying sensitivity maps. Finally, all the reconstructed images were visually scored in terms of the level of artifacts (1: no artifact; 2. minor artifact; 3. significant artifact but diagnostic; 4. severe artifacts but barely diagnostic; 5. non-diagnostic due to extreme artifacts).

Results and Discussion

Figure 1 illustrates the short axis view of stress CMR for one of the datasets. As evident from the images, compared to SS, respiratory phase-specific TSS suppresses motion artifacts; however, its performance is not consistent across all frames, perhaps due to lack of adequate, high-SNR ACS data. In contrast, the images reconstructed using MS yield consistent results with least artifacts. Figure 2 shows the temporal profiles along the dashed lines drawn on the top-left image in Figure 1 and also for an additional dataset (not shown). Results from visual scoring of artifacts by two expert readers are listed in Table 1. The level of artifact in MS (2.1±0.4) was lower than that in SS (3.0±0.8) and TSS (3.0±0.5).

Conclusion

Using data from twelve volunteers, we have demonstrated that multiple sets of sensitivity maps generate superior results compared to a single set of sensitivity maps for exercise stress CMR.

Acknowledgements

This work was partially funded by NIH grant R01HL135489.

References

[1] Uecker, M. et al. Magnetic Resonance in Medicine 71 (3), 990–1001 (2014).

[2] R. Ahmad. et al. Magnetic Resonance in Medicine 74 (5), 1266-1278 (2015).

[3] R. Ahmad. et al. IEEE Transactions on Computational Imaging 1 (4), 220-235 (2015).

Figures

Figure 1. Short axis view of real-time exercise stress cardiac images. Two different frames are shown from the same dataset. In frame (a), both TSS and MS provide significant suppression of artifacts, while in frame (b), only MS reduces artifacts significantly.

Figure 2. (a) Temporal profiles along the dashed lines drawn on the top left image in Figure 1. (b) Temporal profiles for an additional dataset (not shown). The arrows highlight visible artifacts.

Table 1. Scores in terms of the level of artifacts (1: no artifact; 2. minor artifact; 3. significant artifact but diagnostic; 4. severe artifact but barely diagnostic; 5. non-diagnostic due to extreme artifact). The values presented are the average of the scores from two expert readers.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)

4894