Patch tensor decomposition and non-local means filter-based hybrid ASL image denoising

Abstract

BackgroundArterial spin labeling magnetic resonance imaging (ASL MRI) is a noninvasive technique to measure cerebral blood flow (CBF). It is widely used in the study of neurodegenerative diseases. Image denoising is an important step in ASL image processing because the signal-to-noise ratio (SNR) of an ASL CBF perfusion image is very small. New methodWe propose a new ASL image denoising method that exploits patch-based low-rank and sparse tensor decomposition and a non-local means filter. Comparison with existing methodsThe proposed method was compared with two existing ASL denoising methods: component-based noise correction method (CompCor) and low-rank and sparse matrix decomposition-based ASL image denoising method (LS-ASLd). ResultsVarious image quality measures, namely SNR, tSNR and ASL CBF variance, show that the proposed method is more effective than existing ASL denoising methods. The proposed method was used to denoise images from a resting state ASL dataset to compute brain functional connectivity (FC) and images from a task-related ASL dataset to identify brain activation. The results show that the proposed denoising method is more effective to enhance the sensitivity of ASL CBF series when undertaking CBF time series-based FC analysis and task activation detection. ConclusionsAssessment of the performance of the proposed hybrid ASL CBF image denoising method confirms that it is especially well-suited to FC analysis and sensorimotor task analysis.