Alzheimer’s disease is a progressive, neurodegenerative disorder that causes memory loss and cognitive decline. While there has been extensive research in applying deep learning models to Alzheimer’s prediction tasks, these models remain limited by lack of available labeled data, poor generalization across datasets and tasks, and the inability to leverage temporal patterns across varying numbers of input scans. In this study, we adapt state-of-the-art temporal self-supervised learning (SSL) approaches for 3D brain MRI analysis and add novel extensions designed to handle variable-length inputs and learn robust spatial features, resulting in three distinct pre-training strategies. We aggregate four publicly available datasets comprising 3,161 patients for pre-training, and show the performance of our models across multiple Alzheimer’s prediction tasks including diagnosis classification, conversion detection, and future conversion prediction. Importantly, our SSL model implemented with temporal order prediction and contrastive learning outperforms supervised learning on six out of seven downstream tasks. By fine-tuning the same pre-trained backbone with one, two, or three input images, we demonstrate that our model is adaptable to varying numbers of input images and time intervals, as well as generalizable across tasks, highlighting its robust clinical potential. We release our code and model publicly at https://github.com/emilykaczmarek/SSL-AD .