Ahmad Seif Kanaan 1,2 , André Pampel 1 , Kirsten Müller-Vahl 2 , Harald E. Möller 1 1 Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany 2 Department of Psychiatry, Social Psychiatry and Psychotherpay, Hannover Medical School, Hannover, Germany Introduction Discussion Methods ISMRM, Toronto, 2015; World Tourette Congress, London 2015 Results Github repository References DC (1) Test-Retest reliability of 1H-MRS absolute metabolite quantitation with partial volume correction using different segmentation methods ■ The use of water as an internal concentration standard, is a feasible and common techniques for accurate 1H-MRS metabolite quantitation. ■ Large spectroscopic voxels commonly contain a mix- ture of grey and white matter (GM, WM) and cerebro- spinal fluid (CSF). ■ Metabolite signals only arise from GM and WM. ■ GM and WM compartments have different water con- centrations [1] and exhibit different T1/T2 relaxation- time constants for metabolites and water [2,3]. ■ Compartmentation can be taken into account for ab- solute quantitation (eg. Equation 2). ■ Previous work has shown that different segmentation approaches yield different estimates of corrected me- tabolite concentrations [4]. ■ In this study, we investigate the reproducibility of ab- solute metabolite quantitation with internal water referencing while taking partial volume effects into account. ■ We test the effects of different segmentation strategies on the reproducibility of metabolite quantitation in cortical and subcortical regions with varied tissue frac- tion content. Re-localization Accuracy ■ The relocalization procedure yielded Dice coeffi- cients Means and SDs of 0.76 + 0.14 for the ACC voxel, 0.81 + 0.13 for the THA voxel, 0.12 + 0.14 WM voxel. Segmentation Consistency ■ No significant differences were observed in tissue fraction estimates between sessions using the three algorithms. ■ SPM & FSL estimates were similar, however, Freesurfer GM estimates were significantly dif- ferent (20-30% lower in the ACC and 30-40% higher in the THA, 5-10 % lower WM). ■ SPM exhibited the highest consistency (lowest COV%) in tissue fraction estimates between ses- sions. LC-Model Quantitation ■ Five metabolite concentrations were considered reliable: tNAA (NAA +NAAG), tCre (Cr + pCr), tCho (Cho + pCho), mI, Glx(Glu + Gln). ■ No significant differences were observed be- tween scans for LC-Model and tissue fraction cor- rected metabolite concentrations. ■ Mean COV%s were lower than 10% for all reliable metabolites. Quantitation Correction ■ Correction with SPM tissue estimates exhibited the highest reliability between-sessions (lowest COV%) for different anatomical sequences and different voxels. ■ Quantitation correction with SPM decreased between-session variance for Glu/tCho in the thalamic voxel. Increases in variance for other metabolites in different regions were slight and non-significant. [1]. Ernst, T., Kreis, R., & Ross, B. D. (1993). Journal of Magnetic Resonance, Series B, 102(1), 1–8. [2]. Mlynárik, V., et. al. (2001). NMR in Biomedicine, 14(5), 325–331. [3]. Stanisz, G. J. et. al. (2005). Magnetic Resonance in Medicine, 54(3), 507–12. [4]. Gasparovic, et. al. (2006). Magnetic Resonance in Medicine, 55(6), 1219–26. [5]. Gruetter, R. (1993). Magnetic Resonance in Medicine, 29:804-811, [6]. Gruetter, R., et. al. (2000). Magnetic Resonance in Medicine; 43:319-323 [8]. Scott, KT, (2006). MAGNETOM Flash 1/2006; 98–103. [7]. Provencher, SW.(1993). Magn Reson Med ;30(6):672–9 [9]. Gussew, A., et. al . (2012). Magma, 25(5), 321–33. [10]. Christiansen, P., et. al. (1993). Magnetic Resonance Imaging, 11(1), 107–118. [11]. Lee, B.-Y., et.al. (2013). Proc. Intl. Soc. Mag. Reson. Med. 21 (2013) 2033. Data Acquisition ■ Dataset 1 ▶ Subjects: 10 healthy controls (age = 28.3+2.3 ). ▶ System: 3T MAGNETOM Trio (Siemens, Erlangen, Germany). ▶ MPRAGE: TR=1s, TE=2.7ms, FOV=192mm, 256x256 acq. matrix, 1.0mm 3 . ▶ 1H-MRS: Frontal White matter Voxel (WM). PRESS TE= 30ms, TR=5000ms, 3.0mL, 128 supp averages, 16 unsupp acq, 8-channel head coil. ■ Dataset 2 ▶ Subjects: 20 healthy controls (age= 38.9+11.4). ▶ System: 3T MAGNETOM Verio (Siemens, Erlangen, Germany). ▶ MP2RAGE: TR=5s, TE=3.93ms, FOV=192mm, 256x256 acq. matrix, 1.0mm 3 . ▶ 1H-MRS: Anterior Cingulate Cortex (ACC) : PRESS TE= 30ms, TR=3000ms, 6.4 mL, 80 supp acq., 16 unsupp acq., FASTESTMAP shimming  [5,6], AutoAlignHead repositioning [7], 32-channel head coil. ▶ 1H-MRS: Bilateral Thalamus (THA): PRESS TE= 30ms, TR=3000ms, 7.2 mL, 80 supp acq., 16 unsupp acq., FASTESTMAP shimming, AutoAlignHead repositioning. Voxel Registration ■ MRS voxel was mapped onto anatomical space by calculating the transformation matrix from the Siemens Raw Data Format (RDA) file header. ■ Voxel overlap was calculated via the Sørensen–Dice coefficient (DC, equation 1). Tissue Fraction Extraction ■ Three different Segmentation algorithmswere tested: SPM12 NewSegment, FSL FAST, Freesurfer (FSU). ■ GM, WM, CSF tissue percentages were calculated within the limits of the MRS binary mask. Probabilistic maps were binned to make tissue concentrations add up to 100%. Absolute metabolite quantitation ■ Method 1: LC-Model quantitation [8]. ■ Method 2: LC-Model quantitation correction with partial volume correction (Equations 2-3) [9-11]. Relaxation effects were ignored since metabolites have similar T1/T2 times in GM,WM and are approximately accounted for in LC-Model . Statistical Analysis ■ Test-retest reliability was assessed by calculating the coef- ficient of variation for tissue fraction and absolute metabo- liteestimates (COV = SD/Mean). ■ In this study, we examined the reliability of 1-H MRS absolute metabolite quantitation with partial volume correction. ■ We observed that sophisticated and commonly used segmentation algorithms yield different re- gional estimates of tissue fractions. ■ We report that SPM yields the highest consistency in partial volume estimation between-sessions. ■ Although consistency does not imply validity, we observed that quantitation correction with SPM tissue fraction estimates yields the lowest between- session variance across different groups, voxels and anatomical sequences. ■ In comparison with LC-Model quantitation, we report a decrease in variance for key metabolites estimates when considering SPM tissue fractions. ■ To interrogate changes of relevant metabolites in the longitudinal setting in psychiatric or neu- rological disorders, we recommend correction for 1H-MRS partial volume effects, which can be achieved with MRI T1-weighted based segmenta- tions. Acknowledgements ASK is funded by Marie-Curie Initial Training Network TS-EUROTRAIN; FP7-PEOPLE-2012-ITN, GA no 316978. The study was funded in part by the Helmholtz Alliance “ICEMED”. (2) (3) I I 0.71 ; 0.81 ; 1.0 Figure 1. Test-retest reproducibility of voxel locali- zation, tissue fraction estimation, quantitation cor- rection for the ACC Voxel. Figure 2. Test-retest reproducibility of voxel lo- calization, tissue fraction estimation, quantitation correction for the THA Voxel. Figure 3. Test-retest reproducibility of voxel locali- zation, tissue fraction estimation, quantitation cor- rection for the WM Voxel. tCr tCho tNAA mIno Glu Gln Glx tCr tCho tNAA mIno Glu Gln Glx tCr tCho tNAA mIno Glu Gln Glx 0 4 8 12 COV (%) 0 4 8 12 16 COV (%) 10 0 20 30 COV (%) 3 1 Test Scan Re-test Scan GM WM GM WM CSF GM WM CSF Between-Session Variance of tissue fraction estimates Between-Session Variance of absolute metabolite quantitation Between-Session Variance of absolute metabolite quantitation Between-Session Variance of absolute metabolite quantitation Between-Session Variance of tissue fraction estimates Between-Session Variance of tissue fraction estimates 5 15 30 COV (%) 5 15 30 COV (%) 5 15 30 COV (%) SPM FSurfer FSL SPM FSurfer FSL SPM FSurfer FSL SPM LCM FSurfer FSL SPM LCM FSurfer FSL SPM LCM FSurfer FSL Test Scan Re-test Scan Test Scan Re-test Scan 2 SPM FSL FSU SPM FSL FSU SPM FSL FSU 1.0 PPM PPM 1.0 2.0 3.0 4.0 PPM 1.0 2.0 3.0 4.0 2.0 3.0 4.0 View publication stats View publication stats