SPLTRAK Abstract Submission

Number: 1174 Submitted By: Bruce Fischl Last Modified: 11 Jan 05 CNR Comparison of Three Pulse Sequences for Structural MR Brain Imaging Xiao Han1, Jorge Jovicich1, David Salat1, Andre van der Kouwe1, Bradford Dickerson2, Diana Rosas2, Nikos Makris2, Anders Dale3, Bruce Fischl1 1Athinoula A. Martinos Center, Massachusetts General Hospital, Charlestown, MA, 02129, 2Department of Neurology, Massachusetts General Hospital, Charlestown, MA, 02129, 3Department of Neurosciences and Radiology, University of California, San Diego, CA 92093 Objective: Accuracy of quantitative MR imaging-based neuroanatomical study is inherently limited by the acquired image quality such as resolution, distortion, and contrast-to-noise-ratio (CNR). A high-bandwidth multi-echo FLASH (MEF) sequence was recently developed by our group that allows significant reduction in image distortion while maintaining high image SNR [1]. In this work, we further investigate the ability of this new pulse sequence to differentiate multiple brain structures by comparing the image CNR against two commonly used sequences: the single echo FLASH (seFLASH) [2] and the MPRAGE [3]. Methods: Data Acquisition: 12 subjects were scanned at a Siemens 1.5T MR scanner. In each scan session the following data were acquired: two MPRAGE volumes (averaged afterwards after motion-correction) (190 Hz/pixel, TR/TE/TI = 2.73s/3.44ms/1s, flip angle = 7o); two multiple flip angle (30o and 5o) seFLASH volumes (TR = 20ms, TE = 6ms); and two multiple flip angle (30o and 5o) MEF volumes (651 Hz/pixel, TR = 20ms, TE = (1.8+1.82*n)ms, n=0~7) [1]. The acquisition time and image resolution were the same for each pulse sequence. Manual labeling: To evaluate CNR, manual segmentation was performed to generate underlying tissue labeling [4]. Due to time-limitation, manual labeling was generated only for 7 of the 12 MEF datasets, 5 MPRAGE, and 4 seFLASH. CNR computation: We used Mahalanobis distance to evaluate CNR for both averaged MPRAGE images and multi-spectral seFLASH and MEF images: CNR(i,j)= √(μi μj)((Σi + Σj)/2)(μi μj), where μ and Σ represent class mean and covariance matrix respectively. Results & Discussion: We computed CNR across different pairs of major brain structures and Figure 1 shows the results. In general, MEF is significantly better than seFLASH in differentiating all pairs of brain structures. Although the gray/white CNR is slightly lower, MEF is significantly better than MPRAGE in differentiating various subcortical structures. In a separate study (reported in a separate abstract by Jovicich et al.), it is also shown that MEF has significantly less intensity variation across multiple scans of the same subject. The improved CNR, reduced intensity variation, and reduced image distortion render a great advantage of using the MEF pulse sequence in MR imaging-based brain morphometry studies. Conclusions: The MEF sequence gives the best overall acquisition efficiency among all three sequences compared. The good CNR across all brain structures and its ability to allow the calculation of intrinsic MR tissue parameters (T1, T2* and proton-density) [1] enables a wide range of neuroscience studies. Work remains to be done, however, to most effectively use the high-dimensional multi-spectral MEF image data to improve brain segmentation and morphometric analysis. The effect of reduced gray/white CNR (than MPRAGE) on surface-based cortical analysis is also worth further investigation. References & Acknowledgements: This research was supported in part by NCRR (grants P41 SPLTRAK Abstract Submission file:///D:/MGH/Submitted_abstract.htm 2 of 21/11/2005 2:08 PMRR14075, R01 RR16594-01A1, and BIRN U24 RR021382), and the MIND Institute. References[1] Fischl et al. (2004), NeuroImage, 23(1): S69-S84.[2] Frahm et al. (1992), in Magnetic Resonance Imaging, pp. 165-203.[3] Haase et al. (1989), J. Comput Assis. Tomogr., 13:1036-1040.[4] Kennedy et al. (1993), in Functional Neuroimaging, Academic Press.