Brain Age Estimation with Twin Vision Transformer using Hippocampus Information Applicable to Alzheimer Dementia Diagnosis

Brain age estimation using deep learning methods has been associated with detecting and diagnosing neurodegenerative diseases. Alzheimer's disease causes structural and functional changes to the brain throughout a person's life, prompting researchers to use medical imaging as a valuable tool for age estimation. Our study proposes a brain age estimation model that utilizes a twin structural network based on the Visual Transformer (ViT) to solve regression problems. Our input data is T1-weighted magnitude resonance imaging (MRI) of 220 participants taken from the OASIS1 dataset, aged between 45 to 96 years old. The ViT’s inputs are patches from a coronal view that include the hippocampus region (i.e., left and right). Briefly, we construct two paths to investigate brain age estimation for each path and then ensemble the predictions to improve the results. The best mean absolute error on the evaluation data set yielded 3.57 years. In comparison with other works, we acquire a promising MAE. The investigation can help to design an interpretable model for clinical diagnosis.