5023

Reducing T2-related bias in mq-BOLD derived maps of Oxygen Extraction Fraction by 3D acquisition
Stephan Kaczmarz1,2, Jens Goettler1,2, Andreas Hock3, Dimitrios Karampinos4, Claus Zimmer1, Fahmeed Hyder2,5, and Christine Preibisch1,6

1Department of Neuroradiology, Technical University of Munich, Munich, Germany, 2Department of Radiology & Biomedical Imaging, Yale University, New Haven, CT, United States, 3Philips Healthcare, Hamburg, Germany, 4Departement of Radiology, Technical University of Munich, Munich, Germany, 5School of Engineering & Applied Science, Yale University, New Haven, CT, United States, 6Clinic for Neurology, Technical University of Munich, Munich, Germany

Synopsis

Multi-parametric quantitative BOLD (mq-BOLD) measurements have been successfully applied in several studies to assess the vascular oxygenation, however, calculated relative Oxygen Extraction Fraction (rOEF) shows unphysiological systematic elevations. We suspect biased T2-measurements to be the main source of error. Therefore, we present an optimized 3D-GraSE T2-mapping sequence and its evaluation in four stages within phantoms, young healthy controls, elderly controls and internal carotid artery stenosis (ICAS) patients. We found significant T2-decreases, fully consistent with reference-values, and thereby significantly decreased rOEF-values by ~25 % towards physiologically more realistic values. Additional clinical value was demonstrated by detecting focal rOEF-increases in an ICAS-patient.

Purpose

The multi-parametric quantitative BOLD (mq-BOLD) approach is highly promising for assessing vascular oxygenation, revealing relative Oxygen Extraction Fraction (rOEF) maps.1,2 It requires three separate measurements of T2, T2* and relative Cerebral Blood Volume (rCBV) and has been successfully applied in several studies.3-6 However, rOEF-values tend to be systematically elevated due to biased T2-measurements.1 This work therefore focuses on improving T2-mapping for mq-BOLD-imaging by analyzing current method’s errors, optimizing the T2 imaging-sequence and evaluating the new sequence’s performance. Measurements were performed in four stages with phantoms, young healthy controls (YHC), elderly healthy controls (EHC) and internal carotid-artery stenosis (ICAS) patients.

Methods

All measurements were performed on a clinical Philips 3T Ingenia MR-Scanner (Philips Healthcare, Best, Netherlands) using a 16ch head-neck-coil. An overview of the applied imaging sequences and their parameters is given in Fig.1. For T2-mapping, a 2D multi-echo GraSE-sequence (8 echoes, ∆TE$$$\,$$$=$$$\,$$$16$$$\,$$$ms, stimulated echo reduction by even-echo fitting) was originally employed for rOEF-mapping.1 We optimized two 3D multi-echo GraSE sequences for T2-mapping with either 8 echoes and ∆TE$$$\,$$$=$$$\,$$$16$$$\,$$$ms (3D-GraSE-I) or 16 echoes and ∆TE$$$\,$$$=$$$\,$$$10$$$\,$$$ms (3D-GraSE-II). Fit-reliability was assessed by voxelwise fitting-error mapping using the relative sum of fit-deviations averaged per echo. To facilitate artefact recognition, artefact-maps were generated by applying empirical thresholds: Either R2’$$$\,$$$>$$$\,$$$18, FitError(T2)$$$\,$$$>$$$\,$$$10 or FitError(T2*)$$$\,$$$>$$$\,$$$4. For experimental evaluations, first, all T2-mapping sequences were compared to 3D-MESE and single-SE in a phantom consisting of six vessels with different agarose gel and Gd-DTPA concentrations featuring typical cerebral T2-relaxation times. Second, all T2-mapping sequences were compared in-vivo in 10 YHC (age: 28.2±3.9$$$\,$$$y) with additional T2*-mapping for R2’ evaluation $$$\left( R_2’=\frac{1}{T_2^*}-\frac{1}{T_2}\right)$$$. Third, DSC-MRI7 was additionally acquired in seven EHC (age: 69.4±5.0$$$\,$$$y) to evaluate the influence of T2-mapping on rOEF following the mq-BOLD approach: $$rOEF=\frac{R_2’}{c\cdot rCBV}$$ with $$$c=\gamma\cdot \frac{4}{3}\cdot \pi\cdot \Delta\chi\cdot B_0$$$.1 Fourth, T2-mapping induced rOEF-differences were evaluated in 3 patients (age: 65.0±3.6$$$\,$$$y, asymptomatic unilateral high-grade ICAS, NASCET$$$\,$$$>$$$\,$$$70$$$\,$$$%). All in-vivo imaging included FLAIR and MP-RAGE for tissue-segmentation (PGM>0.95 & PWM>0.95). Images were post-processed with SPM128 and custom Matlab9 programs.

Results

Sequence comparisons in the phantom yielded relaxation times between T2$$$\,$$$=$$$\,$$$33-104$$$\,$$$ms for 3D-MESE, which we regarded as reference-values (Fig.2). In comparison, the original 2D-GraSE sequence showed systematic T2-elevations (average: 12.1%), which could be reduced by even echo fitting (6.6%) as reported previously.1 3D-GraSE-I deviated from the reference by 2.4% and 3D-GraSE-II by 1.1%. In YHC, the original 2D-GraSE yielded mean-values of T2,GM$$$\,$$$=$$$\,$$$83.9$$$\,$$$ms and T2,WM$$$\,$$$=$$$\,$$$75.4$$$\,$$$ms (Fig.3). The new 3D-GraSE-I and 3D-GraSE-II revealed T2,GM$$$\,$$$=$$$\,$$$78.2/76.5$$$\,$$$ms, T2,WM$$$\,$$$=$$$\,$$$69.1/66.6$$$\,$$$ms. Resulting 3D-GraSE-II-based R2’ was reduced by -16.9% in GM and -26.0% in WM compared to 2D-GraSE. With additional artefact exclusion, R2’ was decreased in total by -32.5% in GM and -31,5% in WM. In EHC, T2-mapping showed the same trend of decreased T2 and R2’-values by the new 3D-GraSE-II compared to 2D-Grase (Fig.4). Analysis of rOEF with 2D-GraSE based T2-maps yielded: rOEFGM$$$\,$$$=$$$\,$$$0.72$$$\,$$$±$$$\,$$$0.08 and rOEFWM$$$\,$$$=$$$\,$$$1.08$$$\,$$$±$$$\,$$$0.10. By applying 3D-GraSE-II based T2 and removing artefact voxels, rOEF was decreased by ~25% resulting in rOEFGM$$$\,$$$=$$$\,$$$0.53$$$\,$$$±$$$\,$$$0.07 and rOEFWM$$$\,$$$=$$$\,$$$0.81$$$\,$$$±$$$\,$$$0.08. Results in ICAS-patients likewise reveal rOEF-value decreases by 3D-GraSE-II compared to 2D-Grase (Fig.5). Histogram analysis of the artefact-corrected 3D-GraSE-II data demonstrated rOEF-shifts towards lower values and reduction of artefact-related clipped peak values (-74.4%). Interestingly, clear rOEF hyperintensities ipsilateral to the stenosis appear only with 3D-GraSE-II based T2-mapping (Fig.5A).

Discussion

Previously observed T2-elevations with 2D-GraSE could be replicated in all four experimental stages.1 3D-Grase-II yielded best results in phantom and YHC scans, fully consistent with reference values.1 We explain the excellent performance of 3D-GraSE-II by elimination of slice excitation profile related artefacts, higher sampling rate (4 echoes$$$\,\to\,$$$16 echoes) and the longer echo train (128 ms$$$\,\to\,$$$160 ms). Due to decreased T2, we measured lowered R2’-values with 3D-Grase-II in YHC, EHC and ICAS-patients. Consequently, rOEF was significantly decreased in EHC and ICAS-patients yielding physiologically more reasonable rOEF-values. In spite of restrictive tissue segmentation (PGM>0.95 & PWM>0.95), assessment of artefact-maps still suggested strong influence of CSF partial volume effects and iron deposition. Besides clearly improving the reliability of mq-BOLD, the new 3D-GraSE-II sequence additionally seems to increase the sensitivity to detect focal rOEF-changes (Fig.5A).

Conclusion

Our results demonstrate successful implementation of the 3D-GraSE-II sequence, which yields T2-values in agreement with reference measurements within reasonable scan time. Previously observed systematical rOEF-elevations could be significantly reduced. Thereby, the reliability of mq-BOLD was greatly improved. Additionally, 3D-GraSE-II allowed to detect a focal ICAS-related rOEF-increase. The proposed artefact-voxel detection by R2’-maps and fit-errors further improved the method. However, artefact-analysis also highlights remaining bias due to CSF partial-volume-effects, suggesting future CSF-suppression to further improve mq-BOLD based rOEF-imaging. Further validation of improved mq-BOLD derived rOEF could be performed by comparison to a reference measurement.

Acknowledgements

The authors acknowledge support by the Friedrich-Ebert-Stiftung and Dr.-Ing. Leonhard-Lorenz-Stiftung.

References

1: Hirsch NM, Zimmer C & Preibisch C et al (2014). „Technical considerations on the validity of blood oxygenation level-dependent-based MR assessment of vascular deoxygenation.“ NMR Biomed, 27: 853-862.

2: Christen T, Schmiedeskamp H, Zaharchuk G et al (2012). „Measuring brain oxygenation in humans using a multiparametric quantitative blood oxygenation level dependent MRI approach.“ MRM 68(3): 905-911.

3: Preibisch C, Shi K, Kluge A, Lukas M, Wiestler B, Gottler J, Gempt J, RingelF , Jaberi MA, Schlegel J, Meyer B, Zimmer C, Pyka T and Forster S (2017). “Characterizing hypoxia in human glioma: A simultaneous multimodal MRI and PET study.” NMR Biomed 30(11).

4: Toth V, Forschler A, Hirsch NM, den Hollander J, Kooijman H, Gempt J, Ringel F, Schlegel J, Zimmer C & Preibisch C (2013). „MR-based hypoxia measures in human glioma.“ J Neurooncol, 115: 197-207.

5: Gersing AS, Ankenbrank M, Preibisch C et al (2015). „Mapping of cerebral metabolic rate of oxygen using dynamic susceptibility contrast and blood oxygen level dependent MR imaging in acute ischemic stroke.“ Neuroradiology 57(12): 1253-1261.

6: Bouvier J, Detante O, Tahon F, et al (2015). „Reduced CMRO2 and cerebrovascular reserve in patients with severe intracranial arterial stenosis: A combined multiparametric qBOLD oxygenation and BOLD fMRI study.“ Human Brain Mapping 36(2) 695-706.

7: Kluge A, Lukas M, Preibisch C et al (2016). „Analysis of three leakage-correction methods for DSC-based measurement of relative cerebral blood volume with respect to heterogeneity in human gliomas.“ MRI 34(4): 410-421.

8: SPM12: Statistical Parametric Mapping software (SPM12) Version 6225: www.fil.ion.ucl.ac.uk/spm.


9: MATLAB and Statistics Toolbox Release 2013a, The MathWorks, Inc., Natick, Massachusetts, United States.

10: Vinci software, Max-Planck-Institut für neurologische Forschung, Cologne, Germany: http://www.nf.mpg.de/vinci3/. Assessed 09.Nov 2015.

11: Bai R, Koay C, Hutchinson E, Basser PJ (2014). "A framework for accurate determination of the T(2) distribution from multiple echo magnitude MRI images." Journal of magnetic resonance (San Diego, Calif. : 1997) 244: 53-63.

Figures

Figure 1: Overview of MRI sequences within the four experimental stages. Applied sequences are marked in the matrix for each stage. The original 2D-GraSE sequence (orange) and new 3D-GraSE sequences (green) are highlighted. T2* and DSC sequences were applied as described previously.1 In the first stage, 2D and both 3D multi-echo GraSE I&II were compared to reference measurements in a phantom. In the second stage, the impact of 3D-GraSE I&II on R2’ was compared in-vivo for YHC. In the third and fourth stage, DSC-data was additionally acquired to evaluate the influence of 3D-GraSE-II acquisition on rOEF-values of EHC and ICAS-patients.

Figure 2: Stage 1 analysis with comparisons of T2-values within six phantom VOIs. The reference sequence is 3D-MESE. Multiple single-SE were measured for reference validation, single-SE T2 is slightly increased, probably by noise floor effects of magnitude images at long echo-times.11 The original 2D-GraSE sequence was fitted twice, with all echoes and only even echoes. For 3D-GraSE, all echoes were fitted. Resulting T2-values are summarized in A (mean ± SD). Comparison of T2-differences relative to 3D-MESE demonstrate reduced deviations of 2D-GraSE by fitting only even echoes. Average deviations are noted for each sequence. Best accordance was found for 3D-GraSE-II (B).

Figure 3: Stage 2 analysis with comparisons of T2-mapping sequences and their impact on R2’ in YHC. The original 2D-GraSE (orange) and new 3D-GraSE sequences (green) were compared (group mean ± SD) in GM and WM (PWM>0.95 & PWM>0.95) (A). Previously reported T2-values in YHC were replicated with 2D-GraSE.1 With additional T2*-mapping, R2’-values were calculated and compared for 2D-GraSE and 3D-GraSE-II. Impact of artefact exclusion was additionally analyzed for 3D-GraSE-II (A). Data of an exemplary participant (B) showing T2*, 2D-GraSE (orange), 3D-GraSE-II based T2 (green) and derived R2’-maps.10 The artefact-map for 3D-GraSE-II (yellow) corresponds well with elevated R2’-regions (blue arrows).

Figure 4: Stage 3 analysis with comparison of T2-mapping sequences in EHC and their impact on rOEF-values. 2D-GraSE (orange) and 3D-GraSE-II sequences (green) were compared on group level (mean ± SD) in GM and WM (A). With additional T2*- and rCBV-mapping, rOEF-values were calculated and compared for 2D-GraSE and 3D-GraSE-II. Paired scatterplots compare T2, R2’ and rOEF-values in GM and WM for 2D-GraSE vs. 3D-GraSE-II (B). Each point corresponds to one participant’s mean-value. Lines indicate corresponding values for the same participant demonstrating significant parameter reductions by 3D-GraSE-II (p<0.0001). Group-averages are indicated by red lines. Asterisks mark corresponding data from (A).

Figure 5: Stage 4 analysis with exemplary data from a left-sided ICAS-patient. rOEF-maps derived from 2D-GraSE vs. 3D-GraSE-II and artefact-maps are compared for two slices (A). All rOEF-maps are displayed within the same range of values. Note, the focal rOEF hyperintensity ipsilateral to the stenosis was only depicted by 3D-Grase-II rOEF (red arrow). The striatum with high-iron content is clearly visible in artefact-maps and corresponds to maximum rOEF-values (blue arrows). Histograms analysis of this patients’ data underlines the rOEF-decrease by 3D-GraSE-II compared to 2D-GraSE (B). Note that artefact removal clearly reduced the frequency of clipped peak-values at rOEF=1.8 (B:Third column).

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
5023