2021/06/28 by Zalan Fabian, Reinhard Heckel, Fabian, Zalan +3 · 2 citations
Engineering · Medicine · #Advanced MRI Techniques and Applications #Advanced X-ray and CT Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #I.2 #I.4 #Image and Video Processing (eess.IV) #J.3 #Machine Learning (cs.LG) #Medical Imaging Techniques and Applications #Sparse and Compressive Sensing Techniques #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2106.14947
openalex publication_date 2021/06/28 · openalex created_date 2021/07/05 · openalex updated_date 2026/07/28
Deep neural networks have emerged as very successful tools for image\nrestoration and reconstruction tasks. These networks are often trained\nend-to-end to directly reconstruct an image from a noisy or corrupted\nmeasurement of that image. To achieve state-of-the-art performance, training on\nlarge and diverse sets of images is considered critical. However, it is often\ndifficult and/or expensive to collect large amounts of training images.\nInspired by the success of Data Augmentation (DA) for classification problems,\nin this paper, we propose a pipeline for data augmentation for accelerated MRI\nreconstruction and study its effectiveness at reducing the required training\ndata in a variety of settings. Our DA pipeline, MRAugment, is specifically\ndesigned to utilize the invariances present in medical imaging measurements as\nnaive DA strategies that neglect the physics of the problem fail. Through\nextensive studies on multiple datasets we demonstrate that in the low-data\nregime DA prevents overfitting and can match or even surpass the state of the\nart while using significantly fewer training data, whereas in the high-data\nregime it has diminishing returns. Furthermore, our findings show that DA can\nimprove the robustness of the model against various shifts in the test\ndistribution.\n