Dynamic image denoising for voxel-wise quantification with Statistical Parametric Mapping in molecular neuroimaging

Abstract

Purpose: PET and SPECT voxel kinetics are highly noised. To our knowledge, no study has determined the effect of denoising on the ability to detect differences in binding at the voxel level using Statistical Parametric Mapping (SPM).

Methods: In the present study, groups of subject-images with a 10%- and 20%- difference in binding of [123I]iomazenil (IMZ) were simulated. They were denoised with Factor Analysis (FA). Parametric images of binding potential (BPND) were produced with the simplified reference tissue model (SRTM) and the Logan non-invasive graphical analysis (LNIGA) and analyzed using SPM to detect group differences. FA was also applied to [123I]IMZ and [11C]flumazenil (FMZ) clinical images (n = 4) and the variance of BPND was evaluated.

Results: Estimations from FA-denoised simulated images provided a more favorable bias-precision profile in SRTM and LNIGA quantification. Simulated differences were detected in a higher number of voxels when denoised simulated images were used for voxel-wise estimations, compared to quantification on raw simulated images. Variability of voxel-wise binding estimations on denoised clinical SPECT and PET images was also significantly diminished.

Conclusion: In conclusion, noise removal from dynamic brain SPECT and PET images may optimize voxel-wise BPND estimations and detection of biological differences using SPM.

Fig 1. Schematic presentation of the simulation experiment.

Fig 2. Recovery (%) (vertical axis) of voxels, in which a difference in radiotracer binding was simulated, plotted against the level of simulated noise (horizontal axis) and the simulated difference in simulated parameters between the groups of scans (represented by different colors).

Continuous lines represent recovery from FA-denoised images while dashed lines represent raw-image derived recovery. Two subplots (a-b) correspond to the different quantification approaches. (c) An axial slice of a parametric image of T-values derived from SPM analysis of difference of binding of two groups of parametric images produced after voxel-wise application of SRTM for a low- (α = 10) and high- noise (α = 20) level as compared in raw and FA-denoised images. Note that FA-denoising leads not only to an increase in recovery of simulated voxels but in an increase in T-values associated with the recovered voxels.

Fig 3. Average TACs over the voxels in the occipital and thalamic VOI, extracted from raw (RAW-Occ and RAW-Thal respectively) and FA-denoised (FA2c-Occ and FA2c-Thal, respectively) simulated images.

The same TACs are shown for all four levels of noise simulation in comparison to corresponding TACs from the simulated, un-noised dynamic image (SIMUL-Occ and SIMUL-Thal). FA application does not induce any considerable bias in voxel-wise kinetics and markedly diminishes its variability.

Fig 4. Average percent normalized residuals between the true simulated and FA-denoised dynamic simulated image TACs for all time-frames and for the four levels of noise (marked with different colors).

Fig 5. Parametric BP_ND (ml/ml) and CV (%) values of binding parameters obtained from voxel-wise quantification using SRTM and LNIGA on an axial slice on a [¹²³I]IMZ SPECT image from one participant of the study.

Denoising with FA gives equal, if not superior quality parametric images of BP_ND while markedly diminishes its variability.

Fig 6. Parametric BP_ND (ml/ml) and CV values of binding parameters obtained from voxel-wise quantification using SRTM and LNIGA on an axial slice on a [¹¹C]FMZ PET image from one participant of the study.

As with SPECT images of Fig 5, denoising with FA gives equal, if not superior quality parametric images of BP_ND while markedly diminishes its variability.