KUL-VBG

KUL-VBG generates lesion-free T1-weighted MRI images and enables whole-brain parcellation in the presence of large focal brain lesions to support accurate anatomical labeling and downstream neuroimaging analyses.

Key Features:

• Lesion-free image generation: Generates lesion-free T1-weighted MRI images to enable subsequent atlas-based processing and parcellation.
• FreeSurfer compatibility: Provides an automated workflow compatible with FreeSurfer recon-all for whole-brain parcellation.
• Pathology handling: Handles a broad range of focal brain pathologies including large, bilateral, intra- and extra-axial lesions, with or without mass effect.
• Clinical validation: Tested on 10 patients with heterogeneous gliomatous lesions and assessed by expert neuroradiologists showing superior performance to non-VBG (Mann-Whitney U tests: U(6,6) = 33, z = 2.738, P < .010).
• Synthetic cohort validation: Quantitatively validated on a synthetic glioma cohort (n = 100) with mean Dice scores of unilateral VBG = 0.894, bilateral VBG = 0.903, non-VBG = 0.617 (one-way ANOVA, P < .001).
• Lesion characteristic analysis: Evaluates the influence of lesion volume, overlap with FreeSurfer volumes of interest, and distance from those regions on labeling accuracy via linear regression.

Scientific Applications:

• Multimodal registration: Facilitates alignment and integration of different imaging modalities for comprehensive brain studies.
• Group comparisons: Enables accurate comparisons across patient groups by mitigating lesion-induced parcellation errors.
• Anatomical reference: Provides a robust anatomical framework for analyses in datasets with structural abnormalities.
• Connectomics: Supports structural and functional connectomics workflows using FreeSurfer, CAT12, SPM, and Connectome Workbench.

Methodology:

Generates lesion-free T1-weighted MRI images and runs an automated workflow compatible with FreeSurfer recon-all; validation used a synthetic glioma cohort derived from healthy control and patient data with quantitative evaluation by Dice scores, one-way ANOVA, Mann-Whitney U tests, and linear regression on lesion characteristics.