The analysis of fMRI complexity as a biomarker of cognitive decline compared to tau‐PET

Abstract

AbstractBackgroundTau‐PET detects the neurofibrillary tangle pathology and is considered a biomarker of neuronal injury and cognitive decline. However, PET imaging has radiation exposure risks and is costly. Recently, brain entropy mapping based on rs‐fMRI has emerged as a brain functionality biomarker that indicates the complexity of BOLD signals. Various studies have implied that reduced brain entropy values correlate significantly with cognitive decline in Alzheimer’s disease. This study analyzes the effect of fMRI complexity as a biomarker of cognitive decline in AD compared to tau‐PET.MethodWe obtained the dataset for this study from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The dataset consisted of 147 subjects (age = 72.5±7.5, M/F = 60/87, CN/MCI/AD = 95/45/7) with a tau‐PET (tracer: 18F‐AV1451) and two fMRI scans. FMRI images underwent standard processing to generate Multi‐Scale Entropy (MSE) maps. We calculated parametric SUVR values based on cerebellar reference regions from tau‐PET images. To determine the ability of MSE and PET images as a risk factor of cognitive decline in AD, convolutional neural networks were used to classify individuals into CN or MCI/AD groups. Each classifier was trained separately with fMRI complexity images or SUVR images. These two models had the same network structures and hyperparameters. We performed a 5‐fold cross‐validation to evaluate each model.ResultThe classifier using fMRI‐MSE achieved an accuracy equivalent to that using tau‐PET (0.7823 vs. 0.7687). We also calculated the Area Under the ROC curve (AUC) to evaluate sensitivity and specificity and observed that the model using fMRI‐MSE achieved AUC of 0.7022, and that using tau‐PET resulted in AUC of 0.6889.ConclusionThe classification performance showed that use of fMRI‐MSE could potentially be an AD biomarker alternative to tau‐PET. Further studies with multisite, larger datasets are required to validate our preliminary results and provide more conclusive evidence.