|
Activity Number:
|
288
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #307260 |
|
Title:
|
A Comparison of Different Methods for Identifying Outliers in MRS Data
|
|
Author(s):
|
Sandra B. Hall*+ and Mihai Popescu and Anda Popescu and Niaman Nazir and Thomas Malone and Robin Aupperle and Allan Schmitt and JoAnn Lierman and William M. Brooks
|
|
Companies:
|
The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center
|
|
Address:
|
3901 Rainbow Blvd., Kansas City, KS, 66160,
|
|
Keywords:
|
brain imaging ; MRS ; outliers
|
|
Abstract:
|
Magnetic Resonance Spectroscopic Imaging provides a non-invasive quantification of metabolite concentrations throughout the brain. This multi-voxel sampling method, however, yields some voxels with non-representative outlying data that should be identified and excluded since they might bias the average concentrations. On the other hand, if we classify too many data points as outliers then we will under estimate the standard deviation for the concentration of each metabolite. Further, if we fail to accurately identify actual outliers this will result in our overestimating the standard deviation. Both of these problems will lead to an inaccurate representation of the concentrations. Using a dataset from two groups of individuals, one with traumatic brain injury and another without injury we will compare different methods that can be used in order to classify data as outliers.
|
