A major drawback of Magnetic Resonance Imaging (MRI) is its long acquisition time, which lasts commonly about 30-60 minutes per scan. This hinders the clinical usage of MRI, reduces patient comfort, reduces the scanner’s throughput and reduces image quality due to motion related artifacts. The aim of this study was to investigate novel methods for reducing the MRI scan time. Two approaches were developed: the first is applicable for static imaging and the other for dynamic imaging. These approaches utilize sub-Nyquist sampling and adequate reconstruction schemes, involving concepts from the theory of Compressed Sensing (CS) and fast Parallel Imaging (PI) using multi-coil arrays.

First, for accelerating acquisition of static MRI images, a Convolution-based Reconstruction for Parallel Imaging (CORE-PI) method was developed. CORE-PI is based on image reconstruction in the Stationary Wavelet Transform (SWT) domain. It advantageously offers flexible undersampling, a parameter-free implementation, non-iterative computations and short runtimes. The method was validated through experiments with in-vivo data obtained from 7Tesla brain scans performed at Leiden University Medical Center (LUMC). The results demonstrated that CORE-PI outperforms current PI methods such as GRAPPA and l1-SPIRiT. CORE-PI provides reconstructions with a higher accuracy than these methods, and its runtimes are significantly shorter (55%-75% and 20% shorter than the GRAPPA and l1-SPIRiT computations respectively).

Then, an approach for reducing acquisition time of a dynamic MRI was developed and implemented on a temporal series of MR-guided High Intensity Focused Ultrasound (MRgHIFU) images. Currently used MRgHIFU methods provide very limited noninvasive temperature mapping. They cover only 3-5 2D slices within the heated volume, due to their long acquisition time. To accelerate such scans, two CS-based methods were developed. The first method accelerates MRgHIFU using complex-differences-CS and multicoil arrays. The method was validated through retrospective experiments with data obtained from clinical in-vivo human prostate cancer treatments (provided by Insightec). The results showed that the proposed method outperforms the l1-SPIRiT method and the state-of-the-art ‘k-space method’. The proposed method produces reconstructions with errors lower by 33%-38% than these methods. Furthermore, it enables a significant scan time reduction, with possibly up to a 10-fold acceleration. The second method that was developed utilized the recent framework of Fast MRI by Exploiting a Reference scan (FASTMER) method. The suggested method embeds into the CS optimization problem a regularization term related to a-priori knowledge about the similarity between pre-heating and post-heating acquired data. The method was demonstrated on data from a HIFU in-vitro heating experiment performed at Stanford’s Radiological Sciences Lab. The results showed that this method is suitable for reconstruction from highly (4-fold) subsampled k-space data and enables flexible undersampling schemes.

In summary, all the methods proposed here may reduce MR scan time significantly and hence improve treatment safety, increase the scanner’s throughput and reduce patient discomfort.