Francesco Caliva1, Victor Kaiyang Cheng2, Rutwik Shah1, Misung Han1, Sharmila Majumdar1, and Valentina Pedoia1
1University of California San Francisco, San Francisco, CA, United States, 2University of California Berkeley, Berkeley, CA, United States
1University of California San Francisco, San Francisco, CA, United States, 2University of California Berkeley, Berkeley, CA, United States
We found ART (Adversarial Robust Training) can encourage the reconstruction of small, clinically relevant features in MRIs and ultimately increase the diagnostic reliability of under-sampled data.
Figure 1: Proposed image reconstruction framework. During training, an attacker introduces difficult to reconstruct features (δ) to under-sampled images (x): it maximizes a reconstruction error between x and a fully-sampled image (y), in the presence of δ, given a network parametrized by θ. Next, network's parameters (θ) are updated to minimize a robust training loss, which includes a reconstruction error and a robust training term that further penalizes reconstruction errors in the regions including abnormalities.
Figure 4 Abnormality reconstruction improved using our proposed ART strategy (based on the SqueezeNet classifier). Reconstruction of 4x undersampled MRI obtained using A) baseline Unet, B) proposed approach; C) fully-sampled MRI. D) Difference map between A) and B), shows clearer visibility of cartilage lesions and a better preservation of abnormal signal changes.
