High-resolution thalamic FDG PET biomarkers for frontotemporal dementia

Introduction: Frontotemporal Dementia (FTD) comprises a diverse set of neurodegenerative conditions marked by progressive degeneration of the frontal and temporal lobes of the brain, manifesting in varied behavioral, linguistic, and motor disturbances. FDG PET imaging of the thalamus frequently reveals disrupted metabolic patterns indicative of its role in neuronal network degradation in FTD. Thalamic FDG PET biomarkers therefore are promising for FTD diagnosis and for monitoring FTD progression. However, the inherently low spatial resolution of PET complicates the accurate quantification of these images. Our study addresses this limitation by employing: (1) Thalamus Optimized Multi-Atlas Segmentation (THOMAS) parcellation for precise delineation of thalamic subregions using T1-weighted magnetic resonance imaging (MRI) and (2) application of a joint entropy (JE) based deblurring framework that leverages high-resolution MRI to recover the resoluton of FDG PET images. Our previous studies have shown the THOMAS framework, which generates a precise segmentation of thalamic nuclei, to be significantly more sensitive and accurate (larger effect size and AUC values) than FreeSurfer for detecting Alzheimer’s disease (AD) via MRI volumetry. The original contributions of this current study include extending the thalamic nuclei segmentation framework to FTD and leveraging the segmentation to super-resolve metabolic FDG PET scans from FTD patients.

Methods: In our study, we employed THOMAS to segment the thalamic nuclei. The thalamic nuclei masks were utilized to synthesize enhanced T1 MR images, which were then used to provide anatomical prior information based on joint entropy (JE) to a PET image deconvolution framework. Imaging data from FrontoTemporal Lobar Degeneration Neuroimaging Initiative (FTLDNI) database was utilized for this investigation. Our cohort included all participants with available T1 MRI, FDG PET, and clinical dementia rating (CDR) scores (N= 50, age 64.10 ± 7.41, 31 females, 24 Behavioural Variant, 8 Semantic Variant, 6 Progressive Non-Fluent Aphasia, 12 Controls). Outlier data were identified and removed based on the normal distribution criterion. Whole brain segmentation was conducted using FreeSurfer, providing whole thalamus regions and cerebellar references for standardized uptake value ratios (SUVRs). Comparative statistical analyses were then executed.

Results: We found that JE deblurring led to improved visual contrast in the thalamus. Correlation analyses of Clinical Dementia Rating (CDR) with FDG PET imaging revealed that: 1) THOMAS parcellation significantly improved R-squared (R²) values and correlation coefficients over FreeSurfer parcellation; and 2) JE deblurring selectively augmented PET scan clarity in specific regions, potentially increasing the reliability of observed statistical correlations. Notably, in regions such as the medial left area of the thalamus, deblurring resulted in a marked rise in R² and a more distinct negative correlation coefficient, indicating a more pronounced inverse relationship post-deblurring.

Conclusions: Our study demonstrates the efficacy of THOMAS parcellation and JE deblurring in enhancing FDG PET imaging for FTD analysis. By segmenting the thalamus into detailed nuclei and boosting PET image resolution, we observed a notable improvement in the statistical correlation with clinical measures of dementia. Our findings underscore the potential of high-resolution thalamic FDG PET imaging biomarkers in FTD and the role PET resolution recovery can play in enabling improved diagnostic precision and disease monitoring in clinical settings for FTD.

• Download figure
• Open in new tab
• Download powerpoint