An improved non-local means filter for denoising dynamic medical imaging, with application to CT perfusion imaging

Current spatiotemporal Non-Local Means (NLM) filters for dynamic medical images are computationally expensive. The aim of this study is to develop an efficient spatiotemporal denoising framework tailored for dynamic medical images.

The proposed method utilizes a decoupled two-step architecture: a prior-guided spatial fast NLM (FNLM) filter followed by a temporal moving average filter (MAF). Dynamic computed tomography images were used for performance evaluation, and the results were compared with those of FNLM, NLM with a spatiotemporal search window (NLM-ST), partial temporal NLM (PATEN) filter, and dynamic NLM (DNLM).

The proposed method achieved computational time equivalent to the FNLM method, while being 15, 78, and 178 times faster than the NLM-ST, PATEN, and DNLM methods, respectively. Simulation results showed that the proposed method outperformed the other four methods in terms of root-mean-square error across all perfusion maps. Experimental results showed that all perfusion values obtained with the proposed method were lower than those produced by the FNLM method (except for the mean transit time in the suspected infarct core, which was higher) and fell between the values obtained from the other three methods. Moreover, the proposed method outperformed the other four methods in terms of the contrast-to-noise ratio of the cerebral blood flow and cerebral blood volume maps.

This study demonstrates that the integration of existing spatial and temporal filtering components (FNLM + MAF) has the potential to be an efficient and effective denoising solution for dynamic medical images.