FDG-PET imaging to identify brain regions associated with Alzheimer's disease-related TDP-43 proteinopathy: A predictive model using penalized logistic regression analysis

BackgroundTAR DNA-binding protein 43 (TDP-43) proteinopathy contributes to Alzheimer's disease progression, but diagnosis remains autopsy-dependent.ObjectiveThis study aimed to develop a predictive model using FDG-PET imaging to identify brain regions associated with TDP-43 pathology in vivo.MethodsPenalized logistic regression analyzed data from 294 participants and a subset of 159 with Braak and Thal ≥4. Features included FDG standard uptake value ratios (SUVRs) from 123 brain regions and age. Participants were split into training (90%) and testing (10%) datasets, with significant predictors identified by non-zero coefficients at optimal lambda.ResultsThe model achieved 68% accuracy (AUC = 0.70) in 294 participants, highlighting the middle temporal gyrus, parahippocampal gyrus, and hippocampus. In the subset (n = 159), accuracy was 65% (AUC = 0.71), with the medial amygdala and precentral gyrus as key predictors of TDP-43 pathology.ConclusionsThis study suggests a moderate predictive accuracy of FDG-PET to identify brain regions associated with Alzheimer's disease-related TDP-43 proteinopathy.