A paper by UII researchers, “Pyramid Convolutional RNN for MRI Image Reconstruction,” has recently been accepted by IEEE Transactions on Medical Imaging (TMI), a top journal in the field of medical imaging with an impact factor of 10.048.
As a non-invasive approach, magnetic resonance imaging (MRI) has many advantages over other imaging modalities. However, the MRI data acquisition is inherently slow, leading to patient discomfort in clinical practice. One common approach to accelerate MRI data acquisition is to take fewer measurements. Fast and accurate MRI image reconstruction from under-sampled data is thus crucial. In this work, the UII America team introduces a novel deep learning based method, Pyramid Convolutional RNN (PC-RNN), to reconstruct images from multiple scales. Based on the formulation of MRI reconstruction as an inverse problem, the team designed the PC-RNN to learn the features in various scales iteratively. The multi-scale modules learn a coarse-to-fine image reconstruction. Unlike other common reconstruction methods for parallel imaging, PC-RNN does not employ coil sensitive maps for multi-coil data but(?) directly model the multiple coils as multi-channel inputs. The coil compression technique is applied to standardize data with various coil numbers, leading to more efficient training and easier deployment to different types of scanners. The proposed method is one of the winners in the 2019 fastMRI competition.
Figure 1. Examples of reconstruction and error maps of knee images for the single-coil task. All errors are multiplied by 5 for better visualization.
Follow UII on WeChat
© 2019 United Imaging Intelligence.
All Rights Reserved.