Fast methods for generating regional cerebral blood flow parametric images with H215O and dynamic PET

To obtain regional cerebral blood flow (rCBF) and tissue-blood partition coefficient, fitting of the two parameters in the Kety model is conventionally accomplished by nonlinear least squares (NLS) analysis, but NLS requires considerable computation time and, therefore, is impractical on pixel-by-pixel analysis for generating parametric images of these parameters. In this study, we investigated several fast methods for the parametric image generation and compared their statistical reliability and computational efficiency. The methods included linear least squres (LLS), linear weighted least squares (WLS), linear generalized least squares (GLS), linear generalized weighted least squares (GWLS), weighted integration (WI), and model-based clustering method (CAK). H215O dynamic PET with Poisson noise component was simulated using numerical Zubal brain phantom. Bias, deviation, and standard deviation of rCBF and partition coefficient, and computation time in various noise environment was estimated and compared. All the methods were implemented with IDL language. Fast algorithms produced parametric images with similar images quality and statistical reliability. When we used CAK and LLS methods combinedly, computation time was significantly reduced and less than 30 seconds for 128 x 128x 46 images on Pentium III processor. Parametric images of rCBF and partition coefficient with good statistical properties can be generated with short computation time acceptable in clinical environment