German Center for Neurodegenerative Diseases
Positron Emission Tomography
Elena Doering studied cognitive science with a focus on neuroscience, computational linguistics and machine learning. She is now working towards her PhD, where she creates biomarker-based machine learning to estimate risk of Alzheimer's disease development.

Presenter of 1 Presentation

COMPARISON OF STRUCTURAL AND METABOLIC BIOMARKERS FOR BRAIN AGE PREDICTION USING MACHINE LEARNING

Session Type
SYMPOSIUM
Date
Fri, 18.03.2022
Session Time
02:45 PM - 04:45 PM
Room
ONSITE: 133-134
Lecture Time
03:30 PM - 03:45 PM

Abstract

Aims

Brain age (BA) is commonly assessed by predicting chronological age (CA) from neuroimaging data of healthy individuals by means of machine learning. During aging, the adult brain undergoes changes both on the morphological and metabolic level. To date, however, BA remains almost exclusively predicted from structural MRI scans. Here, we compare structural (MRI) and metabolic (18F-FDG-PET) biomarkers of neurodegeneration as potential predictors of BA.

Methods

Matched MRI and 18F-FDG-PET scans of 362 cognitively unimpaired individuals were acquired from the ADNI database (adni.loni.usc.edu). Mean gray matter volume and standardized uptake value ratios were calculated from 216 regions of spatially normalized MRI and 18F-FDG-PET scans, respectively. Regression models were trained to predict BA from MRI or 18F-FDG-PET scans using 70% of the data, while the remaining 30% (n = 110) were used for performance evaluation. Mean absolute error (MAE) and R² were assessed between BA and CA, and subsequently compared across the two neuroimaging modalities. Finally, correlations between MRI- and 18F-FDG-PET-predicted BA and neuropsychological test scores from the same visit were calculated.

Results

BA predicted from MRI- and 18F-FDG-PET showed an MAE (R²) of 3.8 (0.5) and 3.71 years (0.53), respectively, and was weakly correlated between modalities. Higher 18F-FDG-PET-, but not MRI-predicted BA, was associated with worse cognitive function.

Conclusions

We deliver first insights that 18F-FDG-PET is useful in the assessment of BA. The established metabolic BA appears to be more sensitive to differences in cognitive function in cognitively unimpaired individuals compared to structural BA.

Hide