Technical Note Comparative Study of Standard Space and Real Space Analysis of Quantitative MR Brain Data Benjamin S. Aribisala, PhD, * Jiabao He, PhD, and Andrew M. Blamire, PhD Purpose: To compare the robustness of region of interest (ROI) analysis of magnetic resonance imaging (MRI) brain data in real space with analysis in standard space and to test the hypothesis that standard space image analysis introduces more partial volume effect errors compared to analysis of the same dataset in real space. Materials and Methods: Twenty healthy adults with no history or evidence of neurological diseases were recruited; high-resolution T 1 -weighted, quantitative T 1 , and B 0 field-map measurements were collected. Algo- rithms were implemented to perform analysis in real and standard space and used to apply a simple standard ROI template to quantitative T 1 datasets. Regional relaxation values and histograms for both gray and white matter tis- sues classes were then extracted and compared. Results: Regional mean T 1 values for both gray and white matter were significantly lower using real space compared to standard space analysis. Additionally, regional T 1 his- tograms were more compact in real space, with smaller right-sided tails indicating lower partial volume errors compared to standard space analysis. Conclusion: Standard space analysis of quantitative MRI brain data introduces more partial volume effect errors biasing the analysis of quantitative data compared to analysis of the same dataset in real space. Key Words: automatic regional analysis; real space; standard space; partial volume effects J. Magn. Reson. Imaging 2011; 33:1503–1509. V C 2011 Wiley-Liss, Inc. QUANTITATIVE ANALYSIS of magnetic resonance imaging (MRI) brain images is an important focus for image postprocessing, since evaluation of small signal changes may be useful as biomarkers in monitoring disease. Quantification often employs either region of interest (ROI) or pixel-wise analysis (eg, voxel-based morphometry, VBM (1), and statistical parametric mapping, (2)). Pixel-wise methods allow for data ex- ploration with no a priori anatomical hypothesis; how- ever, since the analysis is conducted on a single pixel basis, the intrinsic sensitivity is low. Alternatively, ROI analysis sums pixel values over a wide region, giving higher sensitivity, although a priori anatomical knowledge is required to define the target analysis region. In most clinical research applications, ROI analysis is widely used with manual definition of each ROI on the image to be analyzed (3) or on an accom- panying high-resolution image and subsequently applied to the data to be quantified (4). This approach is time-consuming and may display user bias and poor reproducibility. While interactive ROI analysis is normally carried out on the acquired image (ie, in individual or real space), analysis can be performed in a standard space, where target ROIs need only be defined in a single operation and can subsequently be applied to all datasets (5). The availability of electronic brain atlases (eg, the Brodmann brain in MRIcro, (6), ‘‘WFU_Pickatlas,’’ (7), ‘‘MarsBaR,’’ (8), etc) provide tools for automatic definition of ROI. Target regions within the atlas can be directly applied to any dataset which has been transformed into the standard space (9). The most significant step in this type of analysis is the registration of the individual brain to standard space (eg, Talairach or MNI space) which normalizes variations in brain size and shape between subjects. Registration has most commonly used linear transfor- mations (10) but more recently nonlinear approaches have gained popularity (11). In either case this proc- essing step generally requires spatial smoothing and resampling of the original data matrix (12), which may exaggerate partial volume effects across tissue boundaries, biasing analysis of quantitative parame- ters. An alternative approach is to transform the standard space atlas-based ROI into the real space coordinate system of the subject. This latter method retains the convenience of atlas-based analysis while maintaining the original image data intact and is expected to show smaller processing-related bias. The errors associated with quantitative analysis in real and standard space have not to our knowledge been systematically compared. This study therefore compared real and standard space analysis methods and tested the hypothesis that standard space image Institute of Cellular Medicine and Newcastle Magnetic Resonance Centre, Newcastle University, United Kingdom. Contract grant sponsor: Sir Jules Thorn Charitable Trust. *Address reprint requests to: B.A., Newcastle Magnetic Resonance Centre, Newcastle University, Campus for Ageing and Vitality, New- castle-upon-Tyne, NE4 5PL, UK. E-mail: B.S.Aribisala@newcastle.ac.uk Received December 7, 2010; Accepted February 24, 2011. DOI 10.1002/jmri.22576 View this article online at wileyonlinelibrary.com. JOURNAL OF MAGNETIC RESONANCE IMAGING 33:1503–1509 (2011) CME V C 2011 Wiley-Liss, Inc. 1503