Quantification of local image noise variation in PET images for standardization of noise-dependent analysis metrics. (23rd February 2017)

• Record Type: Journal Article
• Authors: Kueng, R; Driscoll, B; Manser, P; Fix, M K; Stampanoni, M; Keller, H
• Is Part Of: Biomedical physics & engineering express. Volume 3:Number 2(2017)
• Publication Date: 2017-02-23
• Subjects: PET image noise -- spatial variation -- acquisition parameters -- reconstruction parameters; Medical physics -- Periodicals; Biophysics -- Periodicals; Biomedical engineering -- Periodicals; Medical sciences -- Periodicals; 610.153
• DOI: 10.1088/2057-1976/3/2/025007
• Languages: English
• ISSNs: 2057-1976

Abstract

Abstract: Purpose. Analysis metrics such as tumor-to-background (T/B) ratios that rely indirectly on local noise properties of a PET image are known to be highly variable depending on image acquisition and image reconstruction parameters. Less well understood are the effects on local image noise arising from correction algorithms that compensate for photon attenuation and detector sensitivity (attenuation and sensitivity corrections). This work investigates these geometry-dependent local noise characteristics in positron emission tomography (PET) images and their implications for the analysis metrics.Methods. Two different phantoms containing homogeneous radioactivity concentrations were scanned for various initial activities and scan times and then reconstructed with a range of reconstruction parameters. Local noise in the form of a local standard deviation was computed and its spatial variation quantified across the PET imaging volume.Results. The local noise variation across the PET image arising from attenuation and sensitivity corrections can be understood with relatively simple analytical models. For fixed reconstruction parameters, the local standard deviation varied up to 60% in the axial direction and up to 30% in trans-axial (radial) direction. Varying the initial activity concentration and the scan time confirmed that the local noise obeys almost perfect Poisson statistics. As a consequence, depending on the spatial location of the volumes of interest, the T/B metrics can vary more than 30%. In addition, the local standard deviation showed substantial variation up to an order of magnitude (factor 10) for different reconstruction parameters.Conclusion. Quantifying the spatial variation of the geometry-dependent local noise across the imaging volume of a particular PET scanner allows us to develop quantitative correction algorithms for standardizing image analysis using noise-dependent metrics.