Investigation of domain specific pretraining of a Swin transformer to improve Alzheimer's disease classification on three different brain imaging modalities

Early prediction of Alzheimer’s Disease is crucial for optimal patient care. It can be achieved by inspecting suitable brain imaging modalities, namely structural T1 weighted MRI, fluorodeoxyglucose-18 positron emission tomography, and arterial spinal labeling. In this work, we present image-based Alzheimer’s Disease classification using a Swin Transformer model and investigate the effect of pretraining on medical image data using Masked Image Modeling. The pretrained model is then downstreamed to predict the three classes of cognitive normal, mild cognitive impairment, and Alzheimer’s Disease using images of the three modalities above retrieved from the ADNI database. Our results demonstrate the positive effect of pretraining on classification metrics for all modalities, reaching 92.9%. 90.3% and 82.7% ROC-AUC for T1w, FDG-PET, and ASL, respectively. They are competitive in comparison to reported state-of-the-art approaches, in particular on the non-invasively retrieved Arterial Spin Labeling data.