Logo Logo
FAQ
Contact
Switch language to German
Normalisierung von hochauflösenden DTI-Datensätzen zu einem Standardtemplate, Optimierung der Normalisierungsmethoden sowie quantitative und qualitative Analyse der Ergebnisse
Normalisierung von hochauflösenden DTI-Datensätzen zu einem Standardtemplate, Optimierung der Normalisierungsmethoden sowie quantitative und qualitative Analyse der Ergebnisse
Spatial normalization of individual data sets to a common template is a crucial step in many neuroimaging data analysis pipelines. Its accuracy has a profound impact on subsequent image analysis and is very likely to influence final results. Here, we investigate the spatial normalization of individual fractional anisotropy (FA) images from diffusion tensor imaging (DTI) to an FA template using three widely used image-processing software packages: FSL, SPM, and ANTs. We compared normalization results using each software’s default settings and after a step-wise adjustment of optional normalization parameters. 37 FA images from 19 healthy controls and 18 patients with non-lesional focal epilepsy were normalized to an FA template in Montreal Neurological Institute (MNI) space. Normalization results were evaluated qualitatively, using isoline display for visual inspection and quantitatively, calculating voxelwise cross-correlation and absolute difference values between each normalized individual FA and the template image. Average cross-correlation values after FSL normalization ranged from 0.903 with default settings to 0.939 with optimized settings with an average intensity difference of 4.7 to 3:7%, respectively. SPM achieved cross-correlation values from 0.788 to 0.877 and intensity differences from 7.0 to 5:5%. ANTs yielded the best quantitative normalization results with cross-correlation values ranging from 0.953 to 0.976 and intensity differences from 3.5 to 2:9%. Visual inspection showed that these results were achieved by ANTs using much stronger local deformations, at the expense of losing various individual anatomical features. These findings illustrate the significant differences between alternative normalization procedures and the effect of optimizing normalization parameters. It is important to adjust those settings to the specific data used and the specific questions asked to ensure a spatial normalization best suited for the intended subsequent image analyses.
DTI Normalisierung Registrierung, Diffusions-MRT FA, Co-Registrierung Fraktionelle Anisotropie, Fractional Anisotropy Co-Registration, Diffusion-MRI Normalization Registration
Bieler, Nadia
2021
German
Universitätsbibliothek der Ludwig-Maximilians-Universität München
Bieler, Nadia (2021): Normalisierung von hochauflösenden DTI-Datensätzen zu einem Standardtemplate, Optimierung der Normalisierungsmethoden sowie quantitative und qualitative Analyse der Ergebnisse. Dissertation, LMU München: Faculty of Medicine
[thumbnail of Bieler_Nadia.pdf]
Preview
PDF
Bieler_Nadia.pdf

5MB

Abstract

Spatial normalization of individual data sets to a common template is a crucial step in many neuroimaging data analysis pipelines. Its accuracy has a profound impact on subsequent image analysis and is very likely to influence final results. Here, we investigate the spatial normalization of individual fractional anisotropy (FA) images from diffusion tensor imaging (DTI) to an FA template using three widely used image-processing software packages: FSL, SPM, and ANTs. We compared normalization results using each software’s default settings and after a step-wise adjustment of optional normalization parameters. 37 FA images from 19 healthy controls and 18 patients with non-lesional focal epilepsy were normalized to an FA template in Montreal Neurological Institute (MNI) space. Normalization results were evaluated qualitatively, using isoline display for visual inspection and quantitatively, calculating voxelwise cross-correlation and absolute difference values between each normalized individual FA and the template image. Average cross-correlation values after FSL normalization ranged from 0.903 with default settings to 0.939 with optimized settings with an average intensity difference of 4.7 to 3:7%, respectively. SPM achieved cross-correlation values from 0.788 to 0.877 and intensity differences from 7.0 to 5:5%. ANTs yielded the best quantitative normalization results with cross-correlation values ranging from 0.953 to 0.976 and intensity differences from 3.5 to 2:9%. Visual inspection showed that these results were achieved by ANTs using much stronger local deformations, at the expense of losing various individual anatomical features. These findings illustrate the significant differences between alternative normalization procedures and the effect of optimizing normalization parameters. It is important to adjust those settings to the specific data used and the specific questions asked to ensure a spatial normalization best suited for the intended subsequent image analyses.