Advances in MR hardware, pulse sequences, and calibration have made quantitative CMR a reality.¹ Quantitative CMR maps are computed from multiple images with different magnetization preparations that are fit to models of the underlying physiology or tissue relaxation parameter.¹ Signal changes that are not captured by the model will cause errors in the parameter estimate. These errors arise from thermal noise (TN) and physiological fluctuations (PN) from cardiac, respiratory, and hemodynamic motion. Reproducibility of quantitative CMR maps is critical for future clinical adoption and depends on the temporal stability of the signal to noise, or temporal SNR (tSNR). To the best of our knowledge, this is the first study that experimentally examines how the choice of imaging parameters affects tSNR and PN in quantitative CMR.

Data Acquisition: Four healthy volunteers were scanned on a clinical 3T scanner (GE Signa Excite HD) with an 8 channel cardiac coil. Images were acquired using snapshot b-SSFP during mid-diastole at a single mid short-axis slice. For each unique set of image acquisition parameters, twenty images were acquired in two breath-holds. A cardiac CINE was acquired in each subject to determine the exact timing of mid-diastole as well as a noise image without RF to calculate the noise covariance matrix. Data were acquired at various matrix sizes (96/192 x 96/128/192) and accelerations (1/1.33/1.6/2) with GRAPPA and partial Fourier imaging using TR/TE/FA = 3.5 ms/1.5 ms/50⁰, BW = 125 Hz, FOV = 300 mm² and a slice thickness of 10 mm. In a single volunteer, images were acquired at different inversion times (125, 275, 875, 1725, 3625 ms) to determine the impact of magnetization preparation on PN and tSNR.

Noise Analysis: Our analysis of PN and tSNR follow the work of Triantafyllou et. al. in the brain.² The total noise in a CMR image series is modeled as the independent sum of TN and PN. Physiological noise scales proportionally with the MR signal level and tSNR is defined as the ratio of the MR signal to the total noise. By measuring image SNR and tSNR, the ratio of PN and TN can be determined with the following relationship:

$$ \frac{PN}{TN}=\sqrt{\left(\frac{SNR}{tSNR}\right)^2-1} $$

Image Reconstruction: Images were coil combined using an optimum B1 weighted combination³ from which SNR maps were calculated. tSNR was measured as the mean intensity of a pixel or ROI divided by the temporal standard deviation. The Fourier transform scale factor, effective noise bandwidth, and parallel imaging g-factor (max 1.40) were used to correct the SNR maps to make them directly comparable with tSNR.^2,4 The GRAPPA g-factor was computed using a direct method from the kernel weights themselves.⁵ The ratio of physiological noise to thermal noise was subsequently determined from the equation above.

Figure 1 contains a representative SNR, tSNR, and PN/TN map at a 1.6x1.6 mm² in plane resolution. Figure 2 contains plots of average SNR, tSNR, and PN/TN within the left ventricular myocardium ROI as a function of the image matrix size and acceleration factor. SNR consistently decreased as the image resolution (matrix size) and acceleration factor (R) increased except in the case of partial Fourier acquisitions (R=1.3,2), which arrive at the k-space center earlier. In contrast, there was no significant change in tSNR with either the acceleration factor or matrix size. This caused low resolution images (96x96) to be PN dominant (PN/TN > 1) regardless of acceleration factor. However, high resolution images (192x128/192) became SNR starved with increasing acceleration, which lead to TN dominance (PN/TN<1).

Figure 3 shows the signal level, SNR, tSNR, and PN/TN at different post inversion delays in a single volunteer. Both SNR and tSNR decrease as the signal approaches the null point of myocardium and rise afterwards. Near the signal null at 1 sec, tSNR approaches SNR and the images become thermal noise dominant (PN/TN < 1) because PN is proportional to signal strength.

The reproducibility and precision of quantitative CMR is dependent on tSNR rather than image SNR. This study has shown that at coarse resolution (>2.5x2.5 mm²) and low acceleration such as those used in ASL and first-pass perfusion, CMR images are PN dominant. At fine spatial resolution (<1.5x1.5 mm²) and high acceleration such as those used in T1, T2, ECV, and BOLD, CMR images become TN dominant. In the PN dominant scenario, improvements in SNR no longer translate to improvements in tSNR (recently documented in ASL⁶). In these cases, the signal degradation from increasing spatial resolution or acceleration will have a negligible impact on tSNR, which we have observed to be relatively stable.