Reducing Uncertainty in Undersampled MRI Reconstruction With Active Acquisition

Abstract

The goal of MRI reconstruction is to restore a high fidelity image from partially observed measurements. This partial view naturally induces reconstruction uncertainty that can only be reduced by acquiring additional measurements. In this paper, we present a novel method for MRI reconstruction that, at inference time, dynamically selects the measurements to take and iteratively refines the prediction in order to best reduce the reconstruction error and, thus, its uncertainty. We validate our method on a large scale knee MRI dataset, as well as on ImageNet. Results show that (1) our system successfully outperforms active acquisition baselines; (2) our uncertainty estimates correlate with error maps; and (3) our ResNet-based architecture surpasses standard pixel-to-pixel models in the task of MRI reconstruction. The proposed method not only shows high-quality reconstructions but also paves the road towards more applicable solutions for accelerating MRI.

SepUnet: Depthwise Separable Convolution Integrated U-Net For MRI Reconstruction

“…Generally speaking, recent studies including the above mentioned works use a small-scale or private dataset to design/train deep learning-based MRI reconstruction models, which hinders reproducibility and further progress in this important area. Moreover, most researches [10][11][12][13] use synthetic k-space data that is not directly acquired but rather obtained from post-processing of already reconstructed images. Although such research works that used small-scale dataset might provide valuable progress in the field of MRI reconstruction, a larger dataset with raw k-space samples is required to fully realize the potential of deep learning in this area.…”

Towards a general framework for fast and feasible k-space trajectories for MRI based on projection methods

“…More recently, deep learning-based methods for MRI have been proposed in the literature as well [14][15][16][17][18][19]. In addition to providing an improvement in the reconstruction quality, the main advantage observed in these methods is the speed of reconstruction compared to iterative reconstruction methods.…”

MRI Measurement Matrix Learning via Correlation Reweighting

“…More recently, in the spirit of growing success in learning-based data acquisition, there has been significant interest in learning measurement matrix [2,10,34]. Some learning-based measurement matrix designing models [1,40] are capable of improving performance by utilizing data statistical information in k-space, which plays a critical role in the final reconstruction.…”

“…For instances, image (c) and image (d) in under-represented subintervals (easily overlooked) contain abundant medical information, which are equally important to image (b) in disease diagnosis as shown in Fig.1. Recent works [1,2,10,15,40] usually assume that there does not exist the imbalance in measurement example procedure. Due to overlooking the exploration this imbalance, these works reveal poor performance for designing the measurement matrix.…”