The Open-Access Journal for the Basic Principles of Diffusion Theory, Experiment and Application
Reconstructing undersampled MR Images by utilizing
principal-component-analysis-based pattern recognition
Fangrong Zong
1
, Marcel Nogueira d’Eurydice
1
, Petrik Galvosas
1
1
MacDiarmid Institute for Advanced Materials and Nanotechnology,Victoria University of
Wellington, Wellington, New Zealand
Corresponding author: Petrik Galvosas, MacDiarmid Institute for Advanced Materials and Nan-
otechnology, School of Chemical and Physical Sciences, Victoria University of Wellington, PO Box
600, Wellington 6140, New Zealand. E-mail: Petrik.Galvosas@vuw.ac.nz.
Abstract
Compressed sensing technique is a recent framework for signal sampling and recovery. It allows
signal acquisition with less sampling than required by Nyquist-Shannon theorem and reduces data
acquisition time in MRI. When the sampling rate is low, prior knowledge is essential to reconstruct
the missing features. In this paper, a different reconstruction method is proposed by using the principal
component analysis based on pattern recognition. The experiments demonstrate that this method can
reduce aliasing artefacts and achieve a high peak signal-to-noise ratio compared to a compressed
sensing reconstruction.
Keywords:
MRI; Fast imaging; Principal components analysis; Recognition; Compressed sensing;
1 Introduction
Acquisition of MRI in clinical applications may be time-consuming and may lead to reduced
patient throughputs and increased image artefacts due to the patient moving during imaging. Reducing
imaging time is financially beneficial to a total cost and can facilitate more studies as well. One
approach of speeding up the acquisition of MRI is to undersample k-space data. Aliasing artefacts due
to the violation of Nyquist-Shannon rule [1, 2] in the reconstructed magnetic resonance (MR) images
can be reduced by a randomly undersampling format [3, 4], which is part of the Compressed Sensing
(CS) framework. CS techniques have been extensively used in undersampled MRI reconstruction
since 2006 [5–8]. The premise of applying a CS framework is that the MR images are sparse in
a certain orthogonal transformation domain (aka basis) [5, 9]. However, when the object is largely
undersampled, the missing features may not be satisfyingly reconstructed. In this situation, prior
knowledge of similar images can provide constructive information to recover those features, such as
shapes and relative contrast to the surrounding tissues.
Principal component analysis (PCA) is a useful statistical technique that extracting the principal
components from a set of objects (similar images, in our case) [10, 11]. The number of principal
components is less than or equal to the number of images in the database, thus the combination of
© 2014, Fangrong Zong
diffusion-fundamentals.org 22 (2014) 14, pp 1-5
1