Exploiting Similarity in Adjacent Slices for Compressed Sensing MRI Lior Weizman 1 , Ohad Rahamim 1 , Roey Dekel 1 , Yonina C. Eldar 1 , and Dafna Ben-Bashat 2 1 Department of EE, Technion, Haifa, Israel 2 Sourasky Medical Center, Tel-Aviv, Israel Abstract—Due to fundamental characteristics of MRI that limit scan speedup, sub-sampling techniques such as compressed sensing (CS) have been developed for rapid MRI. Current CS MRI approaches utilize sparsity of the image in the wavelet or other transform domains to speed-up acquisition. Another drawback of MRI is its poor signal-to-noise ratio (SNR), which is proportional to the image slice thickness. In this paper, we use the difference between adjacent slices as the sparse domain for CS MRI. We propose to acquire thick MRI slices and to reconstruct the thin slices from the thick slices’ data, utilizing the similarity between adjacent thin slices. The acquisition of thick slices, instead of thin ones, improves the total SNR of the reconstructed image. Experimental results show that the image reconstruction quality of the proposed method outperforms existing CS MRI methods using the same number of measurements. I. I NTRODUCTION Magnetic Resonance Imaging (MRI) is a reliable imag- ing method for diagnosis, evaluation and follow-up of brain pathologies, as well as brain activity. However, the acquisition of a routine brain MRI is a relatively slow process. As such, it causes many difficulties, such as patient discomfort during scanning and blurry images due to patient movements during acquisition. Due to the clinical requirement for high resolution MRI, which necessitates acquisition of many data samples at long scanning times, many approaches for MRI acquisition speed-up have been published. In MRI, data is acquired in the Fourier domain of the image, also known as k-space. Compressed sensing [1], [2] techniques have been applied to MRI to significantly reduce the amount of data required for image reconstruction by under-sampling the k-space [3]. CS allows shorter acquisition time by designing specific sub-sampling patterns of the k- space. Reconstruction of the image from the sub-sampled data is then performed by utilizing sparsity of the image in a certain transform domain. The sampling strategy and the reconstruction method are key elements to achieve high quality images from under-sampled k-space data in CS MRI. Over the past decade, various sampling strategies and re- construction methods have been developed, utilizing a variety of transform domains for image reconstruction with CS. Some methods utilize the sparsity of MRI in the wavelet domain or other spatial transform domain for various applications of MRI [3], [4], [5]. Others speed-up dynamic MRI by utilizing the This work was supported by the Ministry of Science and Technology, Israel Sice 22 Slice 23 Difference image Fig. 1: Two adjacent slices taken from 3D MRI scan (top) and the difference im- age between them (left). Slice thickness is 1mm. It can be seen that the differ- ence image is sparse, thanks to the high similarity be- tween adjacent slices. similarity of adjacent time frames in dynamic MRI [6], [7], [8], [9], [10]. In some MRI applications, a 3D image is generated by the acquisition of tens or hundreds of 2D images, coined as image slices. In many cases, adjacent 2D slices are very similar due to the slow spatial variations of the scanned object. This phenomenon is emphasized in brain MRI, where 2D thin slices are usually acquired. This similarity between adjacent slices and the sparsity of the difference image are shown in Fig. 1. In a recently published paper, Pang et al. [4] utilize the sim- ilarity between adjacent slices for MRI image reconstruction with undersampled k-space data. They designed a sampling scheme that samples 25% of the k-space in some slices, and 1% of the k-space in other slices. The similarity between adjacent slices is then used to estimate 25% of the k-space of the low-sampled slices. CS based reconstruction is then applied to all slices to obtain the entire image. While novel in its basic idea, their method prioritizes some of the slices over the others by non-uniform sampling over the slices. 978-1-4244-7929-0/14/$26.00 ©2014 IEEE 1549