Over the next three years our research project aims to develop a cutting-edge artificial intelligence model trained on the extensive functional MRI data from the UK BioBank. Our self-supervised masked autoencoder approach involves training the model by removing patches from functional brain scans and trying to reconstruct the removed patches. That is, the model will learn to recognize and enhance patterns in brain scans without the need for manual labeling. This strategy ensures that healthcare applications using functional MRI data can automatically be improved by integrating with our model that has learned about the patterns of brain activity latent to the large-scale dataset. This ability to enhance the signal present in brain scans can facilitate better diagnosis and monitoring of neurological conditions, ultimately leading to more effective treatments and improved patient outcomes. This project holds significant potential to positively impact public health by improving the signal-to-noise ratio of the data used for medical applications involving functional brain scans.
