Learning From Oversampling: A Systematic Exploitation of Oversampling to Address Data Scarcity Issues in Deep Learning- Based Magnetic Resonance Image Reconstruction

Ibsa Jalata, Reeshad Khan, Ukash Nakarmi

Abstract: Data acquisitions in Magnetic Resonance Imaging (MRI) are inherently slow due to sequential acquisition protocol. Image reconstruction from under-sampled data is posed as an inverse problem in traditional model-based learning paradigms. Recent data-centric learning frameworks such as deep learning (DL) frameworks are data hungry, and demand a large, labeled training data sets. To address the lack of large training datasets, in MRI reconstructions, researchers approach the problem in two ways: (1) unsupervised method where the model is trained without the presence of fully sampled data. (2) using a method that efficiently use the limited dataset for training purpose. In this paper, we first systematically investigate advantages and limitations of current oversampling methods. Then, we also propose a novel oversampling method and a DL framework that systematically exploits the oversampling technique in the learning process as well as also increase the size of training data set. Essentially, we pose the training data oversampling as a one-to-many mapping function and introduce a new loss function based on similarity metric that can be integrated into a DL framework. Our proposed method not only addresses the training data scarcity in MR image reconstruction and improves reconstruction, but also makes the learned model more robust to different under-sampling techniques

External IDs:dblp:journals/access/JalataKN24