Conference Paper

Automatically Learning Cortical Folding Patterns.

Centre for Intell. Machines, McGill Univ., Montreal, QC, Canada
DOI: 10.1109/ISBI.2009.5193310 Conference: Proceedings of the 2009 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, Boston, MA, USA, June 28 - July 1, 2009
Source: IEEE Xplore

ABSTRACT A data-driven technique is presented for automatically learning cortical folding patterns from MR brain images of different subjects. Cortical patterns are represented in terms of generic scale-invariant image features. Learning automatically identifies a set of features that occur with statistical regularity in appearance and geometry from a large set of MR volume renderings, based on a predescribed anatomical region of interest. A filtering technique is presented for distinguishing between valid cortical features and those likely to arise from incorrect correspondences, based on feature geometry. Expert validation of 100 feature instances shows that 77% correctly identify the same underlying cortical structure in different brains despite high inter-subject variability, and filtering improves the ability to identify the most meaningful patterns.

  • Source
    • "where G(x, σ) is a Gaussian kernel of mean x and variance σ, and σ 0 represents the scale of the original image. The Gaussian scale-space arises as the solution to the heat equation when the image is modeled as a diffusion process (Koenderink, 1984), and can be derived from a 2 M. Toews et al. / NeuroImage xxx (2009) xxx–xxx "
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents feature-based morphometry (FBM), a new fully data-driven technique for discovering patterns of group-related anatomical structure in volumetric imagery. In contrast to most morphometry methods which assume one-to-one correspondence between subjects, FBM explicitly aims to identify distinctive anatomical patterns that may only be present in subsets of subjects, due to disease or anatomical variability. The image is modeled as a collage of generic, localized image features that need not be present in all subjects. Scale-space theory is applied to analyze image features at the characteristic scale of underlying anatomical structures, instead of at arbitrary scales such as global or voxel-level. A probabilistic model describes features in terms of their appearance, geometry, and relationship to subject groups, and is automatically learned from a set of subject images and group labels. Features resulting from learning correspond to group-related anatomical structures that can potentially be used as image biomarkers of disease or as a basis for computer-aided diagnosis. The relationship between features and groups is quantified by the likelihood of feature occurrence within a specific group vs. the rest of the population, and feature significance is quantified in terms of the false discovery rate. Experiments validate FBM clinically in the analysis of normal (NC) and Alzheimer's (AD) brain images using the freely available OASIS database. FBM automatically identifies known structural differences between NC and AD subjects in a fully data-driven fashion, and an equal error classification rate of 0.80 is achieved for subjects aged 60–80 years exhibiting mild AD (CDR = 1).
    NeuroImage 02/2010; 49(3-49):2318-2327. DOI:10.1016/j.neuroimage.2009.10.032 · 6.36 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: The variability in cortical morphology across subjects makes it difficult to develop a general atlas of cortical sulci. In this paper, we present a data-driven technique for automatically learning cortical folding patterns from MR brain images. A local image feature-based model is learned using machine learning techniques, to describe brain images as a collection of independent, co-occurring, distinct, localized image features which may not be present in all subjects. The choice of feature type (SIFT, KLT, Harris-affine) is explored with regards to identifying cortical folding patterns while also uncovering their group-related variability across subjects. The model is built on lateral volume renderings from the ICBM dataset, and applied to hemisphere classification in order to identify patterns of lateralization based on each feature type.
    Medical Computer Vision. Recognition Techniques and Applications in Medical Imaging - International MICCAI Workshop, MCV 2010, Beijing, China, September 20, 2010, Revised Selected Papers; 01/2010


Available from