Addressing The False Negative Problem of Deep Learning MRI Reconstruction Models by Adversarial Attacks and Robust Training

摘要

Deep learning models have been shown to be successful in accelerating MRI reconstruction, over traditional methods. However, it has been observed that these methods tend to miss rare small features, such as meniscal tears, subchondral osteophyte, etc. in musculoskeletal applications. This is a concerning finding as these small and rare features are the particularly relevant in clinical diagnostic settings. Additionally, such potentially dangerous loss of details in the reconstructed images are not reflected by global image fidelity metrics such as mean-square error (MSE) and structural similarity metric (SSIM). In this work, we propose a framework to find the worst-case false negatives by adversarially attacking the trained models and improve the models’ability to reconstruct the small features by robust training.