Exploitation of temporal redundancy in compressed sensing reconstruction of fMRI studies with a prior‐based algorithm (PICCS). Issue 7 (9th June 2015)

• Record Type: Journal Article
• Authors: Chavarrías, C.; Abascal, J. F. P. J.; Montesinos, P.; Desco, M.
• Is Part Of: Medical physics. Volume 42:Issue 7(2015)
• Page Start: 3814
• Page End: 3821
• Publication Date: 2015-06-09
• Subjects: biomedical MRI -- brain -- compressed sensing -- image reconstruction -- image sampling -- matrix algebra -- medical image processing -- neurophysiology -- noise -- redundancy -- sensitivity analysis -- spatiotemporal phenomena -- statistical analysis -- variational techniques; Functional imaging -- MRI: anatomic, functional, spectral, diffusion -- Reconstruction; Involving electronic [emr] or nuclear [nmr] magnetic resonance, e.g. magnetic resonance imaging -- Biological material, e.g. blood, urine; Haemocytometers -- Digital computing or data processing equipment or methods, specially adapted for specific applications -- Image data processing or generation, in general; compressed sensing -- fMRI -- rat -- undersampling -- total variation -- prior; Magnetic resonance imaging -- Medical image noise -- Image reconstruction -- Medical image reconstruction -- Brain -- Optical inspection -- Medical X‐ray imaging -- Encoding; Medical physics -- Periodicals; Medical physics; Geneeskunde; Natuurkunde; Toepassingen; Biophysics; Periodicals; Periodicals; Electronic journals; 610.153
• DOI: 10.1118/1.4921365
• Languages: English
• ISSNs: 0094-2405

Abstract

Abstract : Purpose: Compressed sensing is a technique used to accelerate magnetic resonance imaging (MRI) acquisition without compromising image quality. While it has proven particularly useful in dynamic imaging procedures such as cardiac cine, very few authors have applied it to functional magnetic resonance imaging (fMRI). The purpose of the present study was to check whether the prior image constrained compressed sensing (PICCS) algorithm, which is based on an available prior image, can improve the statistical maps in fMRI better than other strategies that also exploit temporal redundancy. Methods: PICCS was compared to spatiotemporal total variation (TTV) and k‐t FASTER, since they have already demonstrated high performance and robustness in other MRI applications, such as cardiac cine MRI and resting state fMRI, respectively. The prior image for PICCS was the average of all undersampled data. Both PICCS and TTV were solved using the split Bregman formulation. K‐t FASTER algorithm relies on matrix completion to reconstruct the undersampled k ‐spaces. The three algorithms were evaluated using two datasets with high and low signal‐to‐noise ratio (SNR)—BOLD contrast—acquired in a 7 T preclinical MRI scanner and retrospectively undersampled at various rates (i.e., acceleration factors). The authors evaluated their performance in terms of the sensitivity/specificity of BOLD detection through receiver operating characteristic curves and by visual inspection of the statistical maps. Results: With high SNR studies, PICCS performed similarly to the state‐of‐the‐art algorithms TTV and k‐t FASTER and provided consistent BOLD signal at the ROI. In scenarios with low SNR and high acceleration factors, PICCS still provided consistent maps and higher sensitivity/specificity than TTV, whereas k‐t FASTER failed to provide significant maps. Conclusions: The authors performed a comparison between three reconstructions (PICCS, TTV, and k‐t FASTER) that exploit temporal redundancy in fMRI. The prior‐based algorithm, PICCS, preserved BOLD activation and sensitivity/specificity better than TTV and k‐t FASTER in noisy scenarios. The PICCS algorithm can potentially reach an acceleration factor of ×8 and still provide BOLD contrast in the ROI with an area under the curve over 0.99.