Non-invasive Blood Oxygen Saturation from Conventional T2 and native T1 CMR mapping: Improved Accuracy through Physiological Flow-Based Correction

Background: Blood oxygen saturation modulates MRI relaxation times due to the distinct magnetic properties of oxygenated and deoxygenated hemoglobin. Prior models, including the Luz–Meiboom (LM) equation, have leveraged T2 relaxation to estimate oxygen saturation, while synthetic hematocrit (HCT) has been inferred from native T1 [1–4]. However, flow-related artifacts remain a key barrier to accurate in-vivo estimation. We aimed to assess the feasibility and accuracy of deriving fully non-invasive image-based blood oxygenation measurement using standard CMR mapping sequences, validated against simultaneously acquired invasive blood samples during MRI-guided catheterization in pediatric patients.

Methods: Native T1 mapping acquired by MOLLI was used to calculate synthetic HCT. The T2 map was derived from images of a T2-prep bSSFP at two echo times: 0 ms and 55 ms. These data were used to generate saturation maps (SatMap) based on the LM model (Figure 1). Image-based oxygenation of desaturated blood was measured by averaging the values of a ROI on the SatMap in the blood-pool of right ventricle (RV). Reference oxygenations of deoxy-blood in the RV were extracted by the inserted catheter from the pulmonary arteries under the same physiological conditions. Additional morpho-functional metrics, including ventricular volumes, ejection fraction, and main pulmonary artery (MPA) flow, were also measured from the images and used to assess their influences on the image-based SbO₂. Correction factors were derived according to the correlations of the morpho-functional metrics and the error in the blood oxygenation value (ΔSbO₂), defined by the difference between the catheter measurement and the image-based measurement.

Results: We retrospectively included 9 pediatric patients (7 restrictive cardiomyopathies, 1 Uhl’s anomaly, 1 congenitally corrected transposition of the great arteries) who underwent MRI-guided cardiac catheterization with concurrent CMR (Dec 2022–Mar 2025). Among evaluated parameters, peak velocity (Vmax) in the MPA showed the strongest correlation with the error (ΔSbO₂, R²=0.97; Figure 2a). Combining Vmax with pulmonary backflow (BF) further improved correlation (R²=0.98; Figure 2b). Root mean square (RMS) error decreased from 16.8% (uncorrected) to 1.82% with Vmax correction and to 1.45% when both Vmax and BF were included, effectively mitigating flow-related artifacts in the image-based blood oxygenation measurement of RV (Figure 3).

Conclusion: This study demonstrates the feasibility of generating a non-invasive blood oxygen saturation measurement from standard CMR mapping sequences by applying the LM equation and synthetic HCT derived from native T1 values. Importantly, flow-related artifacts were effectively corrected using a physiologically grounded approach based on flow measurements in the MPA. Validation against invasive saturation sampling under matched clinical conditions during MRI-guided catheterization supports the robustness and clinical applicability of this technique.

Figure 1. Luz–Meiboom (L-M) model with subject-dependent variables (blue) and fixed constants (green). Example maps show: (a) T₂ map, (b) native T₁ map used to estimate hematocrit (HCT), and resulting saturation map (c, SatMap) derived from both inputs.
Figure 2. The linear regression (a) between peak velocity in the main pulmonary artery (Vmax MPA) and the error in image-derived blood oxygen saturation (ΔSbO₂), calculated as the difference between catheter-based and image-based values. The 2D regression surface (b) combines Vmax MPA and pulmonary backflow (BF MPA) to better account for ΔSbO₂ variability.
Figure 3. An example of the corresponding SatMap, with a region of interest (ROI) placed within the right ventricular cavity to extract blood values, shows root mean square (RMS) error decreasing from 16.77% without correction, to 1.82% with Vmax correction, and to 1.45% with both Vmax and BF corrections.