NA-MIC/Projects/Structural/Shape Analysis/ShapeStatisticsWithPermTestCorrectionAndFDR

Testing Methods for Statistical Shape Analysis

Description

3 Main correction methods in local statistical shape analysis for the multiple comparison problem exists.

1. Bonferroni correction (simple, worst case, conservative)
2. Minimum statistic using Non-parametric permutation tests (less conservative than Bonferroni, full control of family wise error rate (FWER, Type I error, false positives), still pessimistic)
3. False Discovery Rate (FDR, control of false discovery rate, i.e. control of false positive in only those features that reject Null Hypothesis, less conservative)

Currently the UNC shape analysis pipeline incorporates the permutation test based correction technique. This project will

• extend the shape analysis pipeline to incorporate Bonferroni and FDR
• change the current code to use the ITK statistics framework

Current Status

• permutation tests are working
• both bonferroni and FDR will be rather simple to implement, the bulk of work will be associated with transfering the code into the ITK framework
• File:2006 06 PW StatAnal.ppt

Members

• Martin Styner (UNC)