View source for 2008 Winter Project Week:MRISC

{|
|[[Image:NAMIC-SLC.jpg|thumb|320px|Return to [[2008_Winter_Project_Week]] ]]
|}

__NOTOC__
===Key Investigators===
* MIT CSAIL: Mert R Sabuncu
* BWH: Kilian Pohl

<div style="margin: 20px;">

<div style="width: 27%; float: left; padding-right: 3%;">

<h1>Objective</h1>
This study will explore an approach that utilizes an atlas-based joint registration segmentation framework to automatically classify
magnetic resonance images into healthy and disease subjects. We extend an Expectation Maximization framework (Pohl et al. 2006) that unifies atlas-based segmentation and registration. The geometric transformation (i.e., registration with the atlas) is defined using structure-specific affine parameters that yield a global, one-to-one non-linear deformation. The modes of variation of the transformation parameters are generated using principal component analysis. This provides a robust way to learn the global deformation space that aligns a new brain with the atlas. With  this parametrization, the algorithm then performs a deformation-based classification of the new brain using a Support Vector Machine. We demonstrate the algorithm with a data set that consists of 16 first episode schizophrenics and 17 healthy subjects. The data set included manual labels for the (left and right) Superior Temporal Gyrus (STG),  Hippocampus (HIPP), Amygdala (AMY) and Parahippocampal Gyrus (PHG).

</div>

<div style="width: 27%; float: left; padding-right: 3%;">

<h1>Approach, Plan </h1>
Our plan is to build on the technology described in the reference below. 
</div>

<div style="width: 40%; float: left;">

<h1>Progress</h1>

====Jan 2008 Project Half Week====


</div>

<br style="clear: both;" />

</div>


===Reference===

K. M. Pohl, J. Fisher, W.E.L. Grimson, R. Kikinis, and W.M. Wells. A bayesian model for joint segmentation and registration. NeuroImage, 31(1), pp. 228-239, 2006