Abstract
Parallel magnetic resonance imaging (pMRI) is widely used in clinical routine scans. Undersampled data collected from the independent phased-array coils can reduce acquisition time and accelerate the imaging speed. Therefore, the stress and anxiety of a patient with claustrophobia can be alleviated in a closed bore. Although an individual pMRI technique has been used for recovering missing data, aliasing artifacts and noise still exist and deteriorate image quality. On the other hand, ensemble method as the meta-approach in machine learning provides the theoretical guarantee for designing an accurate and ensembled prediction model and outperforming an individual base estimator. In this thesis, an ensemble method is proposed to improve pMRI by integrating multiple individual base models. Multiple images are reconstructed from multiple base models at first, and then they are combined to suppress aliasing artifacts and noise. In the experimental results, the proposed method shows better predictive performance than that of individual base models. Image quality is quantitatively evaluated and improved on both phantom and in-vivo data. Future work will focus on decreasing the computational cost and complexity of the proposed ensemble method.