Calibration, validation, and sensitivity analysis of a 3D method for the mapping of brain venous oxygenation
Deng Mao1,2, Yang Li1, Peiying Liu1, Shin-Lei Peng3, and Hanzhang Lu1

1Russell H. Morgan Department of Radiology, Johns Hopkins University, Baltimore, MD, United States, 2Graduate School of Biomedical Sciences, Univ of Texas Southwestern Medical Center, Dallas, TX, United States, 3Department of Biomedical Imaging and Radiological Science, China Medical University, Taichung, Taiwan

Synopsis

The present study aimed to further develop and investigate of a non-invasive, efficient and reproducible technique to map brain venous oxygenation in 3D. We first improved the processing pipeline by substracting angiogram to remove arterial content and apply threholdings to eliminate poor fitted and low signal voxels. We then calibrated T2* to oxygenation relationship in vitro using the same technique. In addition, we used hyperoxia challenge to test its sensitivity and combined TRUST MRI method for validation.

Purpose

Knowledge of venous oxygenation (Yv) forms the cornerstone toward quantitative estimation of the brain’s oxygen extraction and consumption rate. However, measurements of these parameters are historically a niche market of Positron Emission Tomography (PET). Recently, a novel MRI pulse sequence was proposed that aims to estimate Yv via quantitative mapping of pure blood T2* in a vessel-specific manner in the presence of tissue partial voluming [1]. Despite the initial proof-of-principle, the validity of the method relative to other approaches and its sensitivity in detecting oxygenation changes are not established. In the present study, we aim to address the following issues: 1) while the previous study included both arteries and veins and did not differentiate high fidelity voxels from low ones, the present work devised an analysis pipeline that reports reliable venous voxels only in the final map; 2) while the previous study used a calibration plot based on literature, this study conducted in vitro experiments to obtain a calibration plot that matches exactly the sequence/parameter of the in vivo study; 3) we examined the sensitivity of the technique to hyperoxia challenge; 4) we validated the results with a more established TRUST MRI method [2], when evaluating global brain venous oxygenation.

Methods

Sequence and processing pipeline: This pulse sequence, T2*-Relaxation-Under-Phase-Contrast (TRUPC), was proposed recently and it is essentially a multi-echo phase-contrast sequence. It first separates out pure blood signal using phase contrast and then acquires different T2* weighted images from multiple gradient-echoes, the mono-exponential fit of which yields blood T2* even in the presence of tissue partial voluming (Figure 1). In this work, we devised the following analysis pipeline (Figure 2a) in order to provide an oxygenation map that only includes high-fidelity venous voxels. Arterial voxels will first be removed via a time-of-flight (TOF) angiogram, leaving only the venous voxels. The venous voxel mask is further refined by applying a threshold on delta R2/R2 of the fitting, weeding out voxels with poor goodness-of-fit. Finally, voxels with low S0 were remove to further eliminate unreliable voxels from the map.

Blood calibration experiment: For the in vitro calibration, fresh bovine blood samples were scanned at 37o using a TRUPC protocol identical to that used in humans, from which we obtained a plot of T2* dependence on both oxygenation (Y) and hematocrit (Hct).

Human experiment: Experiments were performed on a 3T (Philips). 4 subjects (2F, 26±2 yo) were studied under resting-state only but the TRUPC sequence was repeated to test the reproducibility. Another 4 subjects (3F, 25±3 yo) were recruited for the hyperoxia study. The subjects were first scanned with TRUPC under room-air breathing and again under hyperoxia state. The hyperoxic gas is expected to increase blood oxygenation. The imaging parameters for TRUPC were: 4 echoes; TR = 60ms; 1st TE/delta TE = 13ms/14ms; matrix size = 288x288x17; resolution = 0.7x0.7x5mm3; Venc = 9cm/s; flow encoding direction: anterior-posterior; acquisition time = 9:48min. For validation purposes, TRUST MRI was also performed.

Results and discussion

Figure 2b shows the step-by-step outcome of oxygenation map using the proposed pipeline. As one can see, the original map contains both arterial and venous vessels. With the application of angiogram, many of the high-oxygenation (red), arterial voxels are eliminated. However, there is still a considerable number of red-colored voxels remaining. This is because fitting in the presence of large noise/fluctuation tends to result in high T2 values and this was verified by Monte Carlo simulations. Therefore, thresholding on delta R2/R2 and S0 was applied and, with these additional criteria, the final remaining voxels manifested primarily venous oxygenation levels (Figure 2b), consistent with expectation. Coefficient-of-variation(CoV) of the Yv map was 6.2±0.87%, suggesting high reproducibility. Figure 3 shows the results of in vitro blood experiments, showing a strong dependence of T2* on Y but also affected by hematocrit to some extent. With results in Figure 3, venous T2* values can be converted to oxygenation. Figure 4a shows venous oxygenation maps during room air and hyperoxia from a representative subject. As expected, venous oxygenation increases globally when breathing hyperoxic gas. Across all subjects, Yv measured with TRUPC increased by 9.0±2.6% (p<0.01). Figure 4b shows a scatter plot between average Yv measured with TRUPC and global Yv measured with TRUST MRI. A strong correlation can be seen (p<0.0001) between the two measures.

Conclusion

With the TRUPC MRI pulse sequence and the proposed analysis pipeline, it is feasible to obtain quantitative maps of venous oxygenation in the human brain.

Acknowledgements

We would like to thank Hao Huang, PhD, Austin Ouyang, MS, Susumo Mori, PhD and Zhipeng Hou, PhD for their help with 3D image display.

References

[1] Mao & Lu. ISMRM 2015. [2] Lu & Ge. MRM 2008.

Figures

Fitting the blood T2* from TRUPC images.

(a) In vitro T2*-Y calibration curves from blood experiment; (b) Interpolated calibration plane from the curves in (a).

Fig. 3. (a) The processing pipeline for TRUPC. (b) The product oxygenation map after each step in (a). Color bars indicate the oxygenation level in %.

(a) An example of oxygenation maps during room air and hyperoxia. Color bars indicate the oxygenation level in %. (b) Correlation between average venous oxygenation measured by TRUPC and global oxygenation measured by TRUST. Blue dots were measured under room air and Red dots were measured under hyperoxia.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1743