Do Larger Matrix Sizes in Positron Emission Tomography Always Result in Increased Image Noise?

McKeown, C. , Gillen, G., Dempsey, M. and Findlay, C. (2015) Do Larger Matrix Sizes in Positron Emission Tomography Always Result in Increased Image Noise? European Journal of Nuclear Medicine and Molecular Imaging 2015, Hamburg, Germany, 10-14 Oct 2015. pp. 1-924. (doi: 10.1007/s00259-015-3198-z)

Abstract

Aim: As part of a PET image quality assessment it was found that reconstructions with a 256 matrix had lower noise values, as assessed by the Coefficient of Variation (COV), than a 192 matrix - a surprising result. The aim of this study was to determine the combined effect of altering post reconstruction filter width and matrix size on image noise. Materials and Methods: A uniform 68Ge flood phantom was used to assess COVon images produced using a GE Discovery 690 PET-CT system. Post-reconstruction Gaussian filter full width half maximum (FWHM) was varied between 0mm and 10mm. Reconstructed field of view (FOV) was fixed at 700mm with 128, 192 and 256 matrices applied. Other reconstruction parameters were also examined to determine if they affected the relative noise performance of the different matrix sizes - effective iterations were varied between 18 and 540, and reconstructions were repeated with/without Time of Flight (TOF) and with/without Point Spread Function correction (PSF). Results: Relative COV for the three matrix sizes behaved as expected at low filter widths (FWHM 0 - 2mm). The 256 matrix produced the highest COVs and the 128 matrix produced the lowest. From an example reconstruction using TOF +PSF with 54 effective iterations: COV256 = 13.7%, COV192 = 13.1% and COV128 = 11.4%. When the filter width was between 2mm and 4mm, the relative COV of the three matrix sizes became unpredictable - larger matrix sizes did not always produce the largest COV results. For example: FWHM =3mm, COV256 (11.9%) was less than COV192 (12.6%). When the filter width was between 4mm and 8mm, the relative COVs of the three matrix sizes were the reverse of expected i.e. COV128 were highest and COV256 were lowest. For example: FWHM = 5mm, COV256 = 8.9%, COV192= 9.1%. COV128= 10.2%. All matrix sizes produced very similar COVs for FWHM above 8mm. All relative COV results were consistent regardless of the number of iterations or the application of TOF and/or PSF. The relative differences became more pronounced, however, as the number of iterations increased (i.e. at higher overall noise levels). Conclusions: The relationship between image noise and matrix size is not straightforward when post reconstruction filtering is applied. Care is therefore required when different matrix sizes are used clinically.

• Item Type: Conference or Workshop Item
• Additional Information: abstract - available from European Journal of Nuclear Medicine and Molecular Imaging 42(Suppl 1): S392
• Status: Published
• Refereed: Yes
• Glasgow Author(s) Enlighten ID: Findlay, Dr Caroline and Dempsey, Dr Mary and McKeown, Dr Clare and Gillen, Dr Gerry
• Authors: McKeown, C., Gillen, G., Dempsey, M., and Findlay, C.
• College/School: College of Medical Veterinary and Life Sciences > School of Cancer Sciences; College of Medical Veterinary and Life Sciences > School of Medicine, Dentistry & Nursing
• Journal Name: European Journal of Nuclear Medicine and Molecular Imaging
• Publisher: Springer Science and Business Media LLC
• ISSN: 1619-7089