MRI image reconstruction using Deep Learning

Abstract

Recent MRI reconstruction methods like SENSE, GRAPPA, and Compressed Sensing (CS) allow for fast acquisition of high-quality images. However, these methods can remain susceptible to folding artifacts that significantly degrade reconstruction quality. To address this problem, we propose a new approach that combines Graph Neural Networks (GNN) with CS to improve reconstruction while preserving anatomical details. Due to initial results of the proposed method and time constraints, we implemented a hybrid post-processing pipeline using CNN, GNN, and CS. Our results show that using CNN for correction cannot fully address the artifacts caused by acquisition. This highlights that significant quality improvements need to come from better initial reconstruction.