Quantitative evaluation of LDDMM, FreeSurfer, and CARET for cortical surface mapping

NUS Graduate School for Integrative Sciences and Engineering, National University of Singapore, Singapore, Singapore.
NeuroImage (Impact Factor: 6.36). 04/2010; 52(1):131-41. DOI: 10.1016/j.neuroimage.2010.03.085
Source: PubMed


Cortical surface mapping has been widely used to compensate for individual variability of cortical shape and topology in anatomical and functional studies. While many surface mapping methods were proposed based on landmarks, curves, spherical or native cortical coordinates, few studies have extensively and quantitatively evaluated surface mapping methods across different methodologies. In this study we compared five cortical surface mapping algorithms, including large deformation diffeomorphic metric mapping (LDDMM) for curves (LDDMM-curve), for surfaces (LDDMM-surface), multi-manifold LDDMM (MM-LDDMM), FreeSurfer, and CARET, using 40 MRI scans and 10 simulated datasets. We computed curve variation errors and surface alignment consistency for assessing the mapping accuracy of local cortical features (e.g., gyral/sulcal curves and sulcal regions) and the curvature correlation for measuring the mapping accuracy in terms of overall cortical shape. In addition, the simulated datasets facilitated the investigation of mapping error distribution over the cortical surface when the MM-LDDMM, FreeSurfer, and CARET mapping algorithms were applied. Our results revealed that the LDDMM-curve, MM-LDDMM, and CARET approaches best aligned the local curve features with their own curves. The MM-LDDMM approach was also found to be the best in aligning the local regions and cortical folding patterns (e.g., curvature) as compared to the other mapping approaches. The simulation experiment showed that the MM-LDDMM mapping yielded less local and global deformation errors than the CARET and FreeSurfer mappings.

27 Reads
  • Source
    • "Whole brain approaches rely on atlases and mapping to flat or spherical templates. Thus these methods are sensitive to the choice of atlases (7) and distortion in mapping to the templates (8). Most region of interest (ROI) approaches have stemmed generally from whole brain parcellation and can thus be influenced by whole brain data rather than local, i.e., ROI image data. "
    [Show abstract] [Hide abstract]
    ABSTRACT: It has been demonstrated that shape differences in cortical structures may be manifested in neuropsychiatric disorders. Such morphometric differences can be measured by labeled cortical distance mapping (LCDM) which characterizes the morphometry of the laminar cortical mantle of cortical structures. LCDM data consist of signed/labeled distances of gray matter (GM) voxels with respect to GM/white matter (WM) surface. Volumes and other summary measures for each subject and the pooled distances can help determine the morphometric differences between diagnostic groups, however they do not reveal all the morphometric information contained in LCDM distances. To extract more information from LCDM data, censoring of the pooled distances is introduced for each diagnostic group where the range of LCDM distances is partitioned at a fixed increment size; and at each censoring step, the distances not exceeding the censoring distance are kept. Censored LCDM distances inherit the advantages of the pooled distances but also provide information about the location of morphometric differences which cannot be obtained from the pooled distances. However, at each step, the censored distances aggregate, which might confound the results. The influence of data aggregation is investigated with an extensive Monte Carlo simulation analysis and it is demonstrated that this influence is negligible. As an illustrative example, GM of ventral medial prefrontal cortices (VMPFCs) of subjects with major depressive disorder (MDD), subjects at high risk (HR) of MDD, and healthy control (Ctrl) subjects are used. A significant reduction in laminar thickness of the VMPFC in MDD and HR subjects is observed compared to Ctrl subjects. Moreover, the GM LCDM distances (i.e., locations with respect to the GM/WM surface) for which these differences start to occur are determined. The methodology is also applicable to LCDM-based morphometric measures of other cortical structures affected by disease.
    Frontiers in Neurology 10/2013; 4:155. DOI:10.3389/fneur.2013.00155
  • Source
    • "Thus, surface-based approaches have recently received great attention and been applied in brain morphometry for exploring abnormalities [4] [28] [31]. Many related comparisons and surveys have also been presented, as in [23] [29] [38]. "
    Computer Vision and Image Understanding 09/2013; 117(9):1107-1118. · 1.54 Impact Factor
  • Source
    • "Thus, surface-based approaches have recently received great attention and been applied in brain morphometry for exploring abnormalities [4] [28] [31]. Many related comparisons and surveys have also been presented, as in [23] [29] [38]. "
    X Chen · H He · G Zou · X Zhang · X Gu · J Hua ·
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents an improved Euclidean Ricci flow method for spherical parameterization. We subsequently invent a scale space processing built upon Ricci energy to extract robust surface features for accurate surface registration. Since our method is based on the proposed Euclidean Ricci flow, it inherits the properties of Ricci flow such as conformality, robustness and intrinsicalness, facilitating efficient and effective surface mapping. Compared with other surface registration methods using curvature or sulci pattern, our method demonstrates a significant improvement for surface registration. In addition, Ricci energy can capture local differences for surface analysis as shown in the experiments and applications.
    Computer Vision and Image Understanding 09/2013; 117(9):1107-1118. DOI:10.1016/j.cviu.2013.02.010 · 1.54 Impact Factor
Show more