Research Abstracts - 2007

Abstract

We investigate a framework where global properties of structural constellations in medical images, i.e., the sizes and configuration of multiple anatomical units, can be employed for population analysis of anatomical variability. Themethod takes advantage of the fact that cross-subject correspondence of certain structures is relatively well-established. This is in contrast with a majority of today's discriminative analysis studies that rely on very local correspondence and/or employ a large number of features from each subject. Moreover, the representations we use can be easily interpreted in meaningful terms of global anatomy, allowing for the potential use of the analysis for exploring the pathology of neurodegenerative diseases like schizophrenia. In this paper, we demonstrate that with a small number of measurements per subject, one can achieve good separation between schizophrenics and matchedcontrols. Our experiments indicate that the location of certain structures can capture discriminative characterization which may not be available through volumetric measurements. For example, we find that the sagittal position of the parahippocampal gyrus is significantly different between schizophrenics and controls. As an example, we also employ widely-available linear and nonlinear machine learning algorithms (the Fisher linear discriminant and Support Vector Machine) up to %84 classification accuracy.

Introduction

In this paper, we propose a population analysis methodology that takes advantage of the well-established structure based correspondence across individuals. The features that we employ (size and position) are global to each structure,so we are not introducing any further ambiguity to the correspondence problem. We combine the statistical power of multiple regions of interest (ROI) to improve the separation between groups. In other words, by jointly examining subtle differences in the position and size of multiple well-defined ROI's, we believe we can do a good job in discovering differences between two groups while characterizing natural variability within a population. As anexample, we report classification results of a study that uses a data set of 17 first episode schizophrenics and 16 age-matched healthy subjects[1]. We employed the manual label maps for the (left and right) Superior Temporal Gyrus(STG), Hippocampus (HIPP), Amygdala (AMY) and Parahippocampal Gyrus (PHG).

Our experiments indicate that the structure positions could capture discriminative characterization not available in volume measurements. Specifically, we find that the sagittal position of the parahippocampal gyrus is significantly moremedial in schizophrenics, compared to controls and effective disorders. We employ widely-available linear and nonlinear machine learning algorithms (the Fisher linear discriminant and Support Vector Machine) to achieve up to %84 classification accuracy.

Background

Thanks to in-vivo imaging technologies like MRI, the last few decades have witnessed a rapid growth in the amount of research that explores structural and functional abnormalities due to certain pathologies (see e.g.[2]). Studies that investigate such inter-group differences heavily rely on a notion of cross-subject correspondence. In schizophrenia research, for example, structural imaging studies have traditionally focused on showing differences between the sizes of certain brain structures, where cross-subject correspondence is well-established [3]. The popularity of volumetric morphology is also due to the fact that size is an easy-to-interpret feature for clinicians.

Some recent research has proposed automatic discriminative analysis techniques that typically investigate/extract thousands of features from each image. There are two flavors to this approach: voxel-based and deformation-based morphology. The first approach attempts to discover statistical differences in, for example, tissue density at various voxel locations between the groups [4]. This assumes a global spatial normalization of the images. Once voxelwise correspondence is set, various local measurements are compared across subjects. The second approach is to employ a highly nonlinear registration algorithm to align the subjects with a template. A discriminative analysis then can be made based on the deformation fields [5, 6] or both the deformation fields and residual images [7].

Collaborators:

Martha E. Shenton, Ph.D.
Professor 
Department of Psychiatry 
Harvard Medical School
VA Boston Healthcare System, Psychiatry (116A) 
940 Belmont St. Brockton, MA 02401 

Sylvain Bouix, Ph.D.

Psychiatry Neuroimaging Laboratory,
Brigham and Women's Hospital
1249 Boylston St., 3rd Floor,
Boston, MA, 02215.

Acknowledgements:

This researhc is supported by the following grants:

• The National Alliance for Medical Image Analysis
  (NIH NIBIB NAMIC U54-EB005149)
• The Neuroimaging Analysis Center
  (NIH NCRR NAC P41-RR13218)
• The NIH NINDS R01 grant on Computational Modeling of Shape Distributions
  (NIH NINDS R01-NS051826)
• The Morphometry Biomedical Informatics Research Network
  (NIH NCRR mBIRN U24-RR021382)
• The Engineering Research Center for Computer-Integrated Surgical Systems and Technology
  (NSF JHU ERC CISST)

References:

[1]Y Hirayasu et al.. Lower Left Temporal Lobe MRI Volumes in Patients With First-Episode Schizophrenia Compared With Psychotic Patients With First-Episode Affective Disorder and Normal Subjects. In Am J Psychiatry , pp. 1384--1391, 1998.

[2]M.E. Shenton, C.C. Dickey, M. Frumin, R.W. McCarley, A review of MRI findings in schizophrenia. In Schizophrenia Research, pp. 1--52, 2001.

[3]J.C. Pruessner et al., Volumetry of Hippocampus and Amygdala with High-resolution MRI and Three-dimensional Analysis Software: Minimizing the Discrepencies between Laboratories. In Cerebral Cortex}, pp. 433--442, 2000.

[4] J. Ashburner and K.J. Friston, Voxel-Based Morphometry - The Methods. In NeuroImage, pp. 805--821, 2000.

[5] C. Davatzikos et al., A Computerized Approach for Morphological Analysis of the Corpus Callosum. In J of Computer Assisted Tomography, pp. 88--97, 1996.

[6] H.P. Volz, C. Gaser and H. Sauer, Supporting evidence for the model of cognitive dysmetria in schizophrenia—a structural magnetic resonance imaging study using deformation-based morphometry. In Schizophrenia Researchpp. 45--56, 2000.

[7] C. Davatzikos et al.,Voxel-based morphometry using the RAVENS maps: methods and validation using simulated longitudinal atrophy. In NeuroImage, pp. 1361–-1369, 2001.