Optimisation of brain SPET and portability of normal databases

Abstract

Use of a normal database in quantitative regional analysis of brain single-photon emission tomography (SPET) facilitates the detection of functional defects in individual or group studies by accounting for inter-subject variability. Different reconstruction methods and suboptimal attenuation and scatter correction methods can introduce additional variance that will adversely affect such analysis. Similarly, processing differences across different instruments and/or institutions may invalidate the use of external normal databases. The object of this study was to minimise additional variance by comparing reconstructions of a physical phantom with its numerical template so as to optimise processing parameters. Age- and gender-matched normal scans acquired on two different systems were compared using SPM99 after processing with both standard and optimised parameters. For three SPET systems we have optimised parameters for attenuation correction, lower window scatter subtraction, reconstructed pixel size and fanbeam focal length for both filtered back-projection (FBP) and iterative (OSEM) reconstruction. Both attenuation and scatter correction improved accuracy for all systems. For single-iteration Chang attenuation correction the optimum attenuation coefficient (mu) was 0.45-0.85 of the narrow beam value (Nmu) before, and 0.75-0.85 Nmu after, scatter subtraction. For accurately modelled OSEM attenuation correction, optimum mu was 0.6-0.9 Nmu before and 0.9-1.1 Nmu after scatter subtraction. FBP appeared to change in-plane voxel dimensions by about 2% and this was confirmed by line phantom measurements. Improvement in accuracy with scatter subtraction was most marked for the highest spatial resolution system. Optimised processing reduced but did not remove highly significant regional differences between normal databases acquired on two different SPET systems.

Data-Driven Optimization of Brain SPECT Processing for Voxel-Based Statistical Analysis

“…indicate a decision is required. Double-sided rectangles indicate decisions that have previously been optimised [6] using a phantom study. Bold rectangles indicate decisions for which options are tested in this work using repeat processing and statistical analysis of a patient image set.…”

“…The processing was also repeated using an optimised uniform tomographic voxel size [6] that differed from the vendor specified value by 1.5%.…”

“…Both filtered backprojection (FBP) and iterative (OSEM) SPECT reconstruction was performed, with optimized [6] uniform attenuation correction, and with or without optimized [6] lower-window scatter subtraction. Prior to spatial normalization (Snorm), facial activity was edited manually and scalp Fig.…”

Thirty year celebration of the contribution of nuclear medicine physicists in Australia

“…In 2000 he validated fully automatic brain SPET to MR co-registration 46 . He also worked with the WHS physics group on the optimization of brain SPECT and portability of normal databases 47 .…”

Data Based Optimization of Brain SPECT Processing for Voxel-based Statistical Analysis

“…indicate a decision is required. Double sided rectangles indicate decisions that have previously been optimised [4] using a phantom study. Bold rectangles indicate decisions for which options are tested in this work using statistical analysis of a patient image set.…”