Tau Content Prediction Based on Brain Network MLP-att Model

Machine learning is widely used in Alzheimer's disease (AD) research. Currently, brain imaging to identify tau protein content can help distinguish between cognitively normal older adults (CN), mild cognitive impairment (MCI), and AD. Imaging studies of tau protein have mainly used positron emission computed tomography (PET). However, PET testing is costly and causes radiation damage. It has been found that the accumulation and propagation of tau protein are closely related to the structure of brain networks. In this paper, we used rs-fMRI to calculate Pearson correlation coefficients of each brain region to obtain static functional connectivity matrices and modeled them using multilayer perceptron neural networks based on attention mechanism (MLP-att). Regression analysis of tau protein content using neural networks on the ADNI dataset had a RMSE of 105.8 and a MAE of 79.0, outperforming traditional machine learning algorithms.