We present a CNN model that assesses diagnostic quality of MRIs in terms of noise, rigid motion and peristaltic motion artifacts. Our multi-task model consists of a set of shared convolutional layers followed by three branches, one for each artifact. We use dual-task training for two branches. We then utilize transfer learning for the third branch for which training data is scarce. We show that multi-task training improves the generalizability of the model, and transfer learning significantly improves the performance of the branch under data scarcity. Our model is deployed in clinically to provide warnings and artifact-specific solutions to technologists.
This abstract and the presentation materials are available to members only; a login is required.