Introduction

Blood oxygen (O₂) saturation is routinely measured using invasive cardiac catheterization in the clinical management of patients with congenital heart disease (CHD). Non-invasive T₂-based MRI blood oximetry has demonstrated promising results in research settings using the Luz-Meiboom (LM) model^1-3, where O₂ saturation is approximated through the reduced blood T₂ associated with increased deoxyhemoglobin. However, improvements in accuracy are needed for this approach to become established clinically. A multiple T₂ map approach was recently proposed to improve the accuracy of O₂ saturation measurements by estimating the 3 unknown nuisance parameters in the LM model⁴: T₂O (T₂ of fully-oxygenated blood), τ_ex (water proton exchange time between erythrocytes and plasma), and α (dimensionless susceptibility parameter). While this technique improved venous O₂ saturation measurements, it relied on a pulse oximeter to calibrate nuisance parameters from arterial blood. However, since τ_ex and α describe the decrease in T₂ caused by deoxyhemoglobin, their estimation at high %O₂ may be inaccurate. Therefore, the purpose of this study was to optimize the measurement of T₂-based blood O₂ saturation, and to validate using MR-guided cardiac catheterization in patients with CHD.

Methods

Simulations
Figure 1A shows the LM model for a range of O₂ saturations. The nuisance parameters were varied to evaluate their effects on the shape of the model curve for arterial blood (95% O₂ saturation, Fig. 1B) and venous blood (60% O₂ saturation, Fig. 1C). While the T₂O can be accurately differentiated at high O₂ saturation, τ_ex and α are insensitive to τ₁₈₀ since the effects of deoxyhemoglobin are negligible (Fig. 1B). Fig 1C shows that at lower O₂ saturations, T₂O and τex are more sensitive to the chosen τ₁₈₀, and suggests that T₂O estimation may improve with a shorter τ₁₈₀, while longer τ₁₈₀ are more sensitive to τ_ex.

Based on these simulations, an alternative O₂ measurement approach is proposed: 1) estimate T₂O from arterial blood, 2) τ_ex and α from venous blood, and 3) extend the τ₁₈₀ range for the four T₂ maps to 5, 8, 12, and 24 ms for improved estimation of T₂O.

MR Imaging
The O₂ mapping sequence was implemented on a 1.5T Ingenia (Philips Healthcare, Best, NL). Four patients with CHD who were scheduled to undergo MR-guided diagnostic cardiac catheterization^5-7 were scanned with the original and proposed mapping schemes, immediately followed by the catheterization procedure where O₂ saturation was measured throughout the heart. Images were acquired using a 2D short-axis single-shot bSSFP (4-chamber for single-ventricle physiology). Other scan parameters were: 70⁰ flip angle, SENSE=2, TE/TR=0.9/1.8ms, 5s recovery between T₂-weighted images, 2x2mm resolution, slice thickness = 8mm, ECG triggered to end diastole, scan time per O₂ map = 2min. For the original scheme, the max number of T₂-prep refocusing pulses for each map was [12, 12, 8, 8]⁴. The proposed short τ₁₈₀ scheme requires additional pulses to adequately sample T₂ decay and used a max number of refocusing pulses of [40, 24, 16, 12].

O₂ mapping procedure
O₂ mapping was performed in MATLAB after correcting the effective TE for the refocusing pulse duration in each T₂-prep module⁸ and applying motion correction using affine image registration. For the original approach, 3 nuisance parameters were estimated using constrained non-linear least square fit of arterial blood with a constant O₂ saturation (from pulse oximeter)⁴. For the proposed approach, all 4 unknown parameters were estimated, and the nuisance parameters were taken from the appropriate blood pool according to the simulations. The average nuisance parameters across subjects were used to calculate the final oxygen maps.

Results

Figure 2 shows the LM model with proposed τ₁₈₀ sampling times (A) and an example of the improved fit that they can provide (B) due to the higher T₂ of blood (C). Monte Carlo simulations were performed to fully evaluate the improved performance of the proposed τ₁₈₀ scheme, which demonstrated reduced mean square error (MSE) for all unknown parameters (Fig. 3).
Figure 4 shows an O₂ map acquired with the proposed approach, showing distinct arterial and venous blood pools that are consistent across imaging planes. The saturation values from the proposed approach agreed better with MR-guided catheterization, providing improved accuracy and reduced bias compared to the original approach (Fig. 5). Across all subjects, the mean ± SD of nuisance parameters (T₂O, τ_ex, α) were 265.9 ± 79 ms, 7.1 ± 3.1 ms, 0.556 ± 0.024 with the original approach, and 196 ± 13.5 ms, 5.5 ± 3.7 ms, and 0.547 ± 0.005 with the optimized technique.

Discussion

The proposed O₂ mapping approach provided very good agreement with catheter-based blood O₂ saturation measurements made under the same anesthetic conditions. Simulations of the LM model aided in optimizing the T₂-prep refocusing intervals and were essential in the selection of blood pool for nuisance parameter learning. The optimized technique also enabled accurate measurements without pulse oximeter calibration, which provides freedom in the choice of the examined blood pool, as the imaging slice is no longer required to contain arterial blood that is assumed to match the pulse oximeter. In conclusion, we have demonstrated improved accuracy of MR-based blood oximetry using optimized multi-parametric estimation of the LM model parameters that matches closely with catheter-based measurements.

Acknowledgements

No acknowledgement found.