Analysis of Outliers Effects in Voxel-Based Morphometry by means of Virtual Phantoms
- Cite this paper as:
- Nocchi F., Franchin T., Genovese E., Longo D., Fariello G., Cannatà V. (2008) Analysis of Outliers Effects in Voxel-Based Morphometry by means of Virtual Phantoms. In: Katashev A., Dekhtyar Y., Spigulis J. (eds) 14th Nordic-Baltic Conference on Biomedical Engineering and Medical Physics. IFMBE Proceedings, vol 20. Springer, Berlin, Heidelberg
Voxel-Based Morphometry (VBM) is a technique for analyzing inter-group neuroanatomic differences, frequently used to study a number of neurologic diseases. An important limit to the diffusion of VBM is related to the length of the recruiting process, due to the need of homogeneous and statistically significant samples. Therefore, quite often the need to search for statistically significant differences between small samples arises. The opportunity of analyzing small samples with VBM should be carefully considered. In each sample there could be one or more subjects with atypical local anatomic characteristics (outliers) that are not identifiable a priori, thus leading to erroneous inferences about the specific structural features correlated to the studied pathology. It follows the need to evaluate the limits within which including a certain number of outliers can be accepted when relying on small samples. The robustness of VBM performed within SPM2 with respect to the inclusion of outliers was studied by implementing in MatLab sets of virtual phantoms with predetermined characteristics. A matrix of voxels cubes was superimposed on the preprocessed gray matter image of each scanned subject. Each cube has a uniform gray level, while its intensity distribution within each group is Gaussian with controlled group mean and variance. Variance and mean difference were chosen according to data from two ongoing clinical studies. The model implemented shows that applying VBM to small samples could lead to erroneous inferences. The minimum number of subjects should be evaluated according to specific variance and group mean difference values of the experimental samples. The implementation of virtual phantoms enabled the quantitative evaluation of outliers effects when performing VBM on small samples. This approach provides the clinicians with a support tool for evaluating the reliability of the results yielded by VBM analyses and for planning the recruiting process.
KeywordsVBM outliers phantoms SPM MRI
Unable to display preview. Download preview PDF.