PET image filtering

Reducing image noise by spatial filtering

Computation of parametric image the noise in dynamic PET image may need to be reduced. Filtering may also allow patient dose reduction (Axelsson & Sörensen, 2013). There are different methods for producing smoother images:

1. Image reconstruction

PET image reconstruction parameters can be modified, either by increasing the level of filtering, or by changing the filter type. FBP reconstruction gives the same result whether filtering is done in the reconstruction or after reconstruction. Iterative reconstruction methods may give variable results depending on the filter applied during the reconstruction. Measured PSF can be applied in the reconstruction to reduce both noise and partial volume effect.

2. Image thresholding

PET images can be thresholded to remove reconstruction artifacts, especially outside the body. Several tools can be used for thresholding dynamic and static images, including our own tools imgthrs and imgcutof for ECAT, Analyze, and NIfTI image format.

3. Image smoothing

Traditional image filtering methods, not related to PET, do not take advantage of the increased signal-to-noise ratio (SNR) of the entire time series. A few methods that apply the time information have been proposed, for example HYPR method (Christian et al., 2010), flexible segmentation method (Bentourkia, 2001), factor analysis (Tsartsalis et el. 2018), and spectral analysis (Veronese et al., 2018). Currently there are no software in TPC for using these methods, although some preliminary programs for testing purposes are available (imgfsegm, imgdysmo).

Gaussian smoothing can be applied using program imgfiltg. Average and SD images from several spatially normalized images can be computed using ecatavg.

Edge-preserving smoothing can be done using for example median filter, bilateral filter (Hofheinz et al., 2011), or non-local means (NLM) denoising (Dutta et al., 2013).

Based on simulations, Huang et al (2003) suggested that noise level in parametric images could be reduced using total variance (TV) de-noising.

Volume-based (3D) smoothing in the brain increases partial volume effects. Cortical surface-based smoothing can reduce the bias and inter-subject variance (Greve et al., 2014). Denoising, partial volume correction, and image segmentation are dependent on each other, and a joint solution can provide better results than performing each step separately (Xu et al., 2018).


See also:



References:

Axelsson J, Sörensen J. The 2D Hotelling filter - a quantitative noise-reducing principal-component filter for dynamic PET data, with applications in patient dose reduction. BMC Med Phys. 2013; 13:1. doi: 10.1186/1756-6649-13-1.

Bian Z, Huang J, Ma J, Lu L, Niu S, Zeng D, Feng Q, Chen W. Dynamic positron emission tomography image restoration via a kinetics-induced bilateral filter. PLoS One 2014; 9(2): e89282. doi: 10.1371/journal.pone.0089282.

Bentourkia M. A flexible image segmentation prior to parametric estimation. Comput Med Imaging Graph. 2001; 25: 501-506. doi: 10.1016/S0895-6111(01)00016-7.

Christian BT, Vandehey NT, Floberg JM, Mistretta CA. Dynamic PET denoising with HYPR processing. J Nucl Med. 2010; 51(7): 1147-1154. doi: 10.2967/jnumed.109.073999.

Dutta J, Leahy RM, Li Q. Non-local means denoising of dynamic PET images. PLoS ONE 2013; 8(12): e81390. 10.1371/journal.pone.0081390.

Floberg JM, Holden JE. Nonlinear spatio-temporal filtering of dynamic PET data using a four-dimensional Gaussian filter and expectation-maximization deconvolution. Phys Med Biol. 2013; 58: 1151-1168. doi: 10.1088/0031-9155/58/4/1151.

Greve DN, Svarer C, Fisher PM, Feng L, Hansen AE, Baare W, Rosen B, Fischl B, Knudsen GM. Cortical surface-based analysis reduces bias and variance in kinetic modeling of brain PET data. Neuroimage 2014; 92: 225-236. 10.1016/j.neuroimage.2013.12.021.

Herholz K. Non-stationary spatial filtering and accelerated curve fitting for parametric imaging with dynamic PET. Eur J Nucl Med. 1988; 14: 477-484. doi: 10.1007/BF00252392.

Jomaa H, Mabrouk R, Khlifa N, Morain-Nicolier F. Denoising of dynamic PET images using a multi-scale transform and non-local means filter. Biomed Signal Proces. 2018; 41: 69-80. doi: 10.1016/j.bspc.2018.05.029.

Kamasak ME. Effects of spatial regularization on kinetic parameter estimation for dynamic PET. Biomed Signal Processing Control. 2014; 9: 6-13. doi: 10.1016/j.bspc.2013.08.011.

Liu H, Wang K, Tian J. Postreconstruction filtering of 3D PET images by using weighted higher-order singular value decomposition. BioMed Eng Online 2016; 15:102. doi: 10.1186/s12938-016-0221-y.

Svensson PE, Olsson J, Engbrant F, Bengtsson E, Razifar P. Characterization and reduction of noise in dynamic PET data using masked volumewise principal component analysis. J Nucl Med Technol. 2011; 39(1): 27-34. doi: 10.2967/jnmt.110.077347.

Tauber C, Stute S, Chau M, Spiteri P, Chalon S, Guilloteau D, Buvat I. Spatio-temporal diffusion of dynamic PET images. Phys Med Biol. 2011; 56: 6583-6596. doi: 10.1088/0031-9155/56/20/004.

Tsartsalis S, Tournier BB, Graf CE, Ginovart N, Ibáñez V, Millet P. Dynamic image denoising for voxel-wise quantification with Statistical Parametric Mapping in molecular neuroimaging. PLoS One 2018; 13(9): e0203589. doi: 10.1371/journal.pone.0203589.

Zhou Y, Huang SC, Bergsneider M, Wong DF. Improved parametric image generation using spatial-temporal analysis of dynamic PET studies. Neuroimage 2002; 15(3): 697-707. doi: 10.1006/nimg.2001.1021.



Tags: , , ,


Updated at: 2019-12-01
Created at: 2013-10-24
Written by: Vesa Oikonen