Extracting Features of T2 Distributions from Magnetic Resonance Images of Human Brain Using Gamma Variate-fitting

• Activity Number: 27
• Type: Contributed
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Angshuman Saha*+ and Sudeshna Adak and Tandon Reeti and John Schenck and Earl Zimmerman
• Companies: General Electric Global Research Center and General Electric Global Research Center and General Electric Global Research Center and General Electric Global Research Center and Albany Medical Center
• Keywords: neuroimaging ; Alzheimer's disease ; histogram shape features ; parametric smoothing ; classification

Abstract

Excessive iron deposition and increased atrophy in the brain are known to be hallmarks of Alzheimer's disease (AD). Such anomalies leave specific signatures in T2 images from a 3T magnetic resonance scan: excess iron reduces T2 while atrophy filled with cerebrospinal fluid (CSF) increases T2. Thus, compared to controls, T2 distributions for Ads show thicker lower tails (due to iron) and thicker upper tails (due to CSF). These differences can be used to diagnose and monitor AD. In the neuroimaging literature, smoothing T2 distributions with gamma variate curves have been suggested for visual appeal. We extract subject specific parameter estimates from a gamma variate fit of the T2 distribution, which are then used to discriminate Ads from controls. We find that fitting a single gamma to the entire histogram is inadequate. We have developed a method for fitting a three component Exponential-Gamma-Exponential model to describe the lower tail (iron), mid section and upper tail (CSF) that provides much better fit and clinically meaningful parameters. We also show that, some of these T2 distribution shape parameters outperform the more traditional volumetric features.