Purpose: Spatial normalization is an essential step in in brain parcellation for connectomic analysis; however, normalization can be confounded by brain lesions. Cost function masking (CFM) is a popular approach for overcoming this difficulty, although newer, higher-degree-of-freedom normalization methodsâ€™ relative performance with or without CFM in lesioned brains is unknown. This study compared three normalization methods with and without CFM, and determined their impact on functional connectomic measures in glioma patients.
Methods: Thirteen patients with gliomas were included in this study. Three-dimensional T1-weighted brain MRIs were used for normalization, using SPM12 with six
methods: original, default, and DARTEL, each with and without CFM. CFM was performed by masking out the tumor ROI, generated by a deep learning algorithm using standard clinical MR images. Edge-highlighted normalized images and ICBM152 T1 template were overlaid and scored by two neuroimaging scientists. Whole-brain connectomic measures, including local and global efficiency, clustering coefficient, betweenness centrality, characteristic path length, modularity and assortativity, were then calculated from the resting-state fMRI data and compared between normalization methods.
Results: The highest performance was achieve by DARTEL without CFM which scored significantly higher than the original method (p=0.027) and marginally significantly higher than the most-used normalization method of original with CFM (p=0.075) using Tukeyâ€™s range test. When using the original and the improved default normalization, CFM improved the brain normalization. Significant to marginally significant (p<0.08) differences in local efficiency, modularity, and clustering coefficient were found when the comparing the original vs. more advanced (default and DARTEL) normalization methods using pairwise t-test without multiple comparison correction, with modularity being the most sensitive to the normalization methods used.
Conclusion: Improvements in normalization methods with DARTEL has reduced the need for CFM in lesioned brain. Using normalization methods that have inferior performance can result in differences in graphical measures from the functional connectomics.