Iterative method for the calculation of parametric images from dynamic brain PET images

1443 Objectives: Image noise in dynamic PET image series may cause dataanomaly in temporal dimension that will affect the parametric imagingprocess using Patlak fitting. Our work is to propose an iterative method fordynamic PET parametric image calculation, which is effective for eliminatingthe affection of data anomaly in temporal dimension and provide a morereliable result of parametric image calculation. Methods: Patient Data Twopatients have been enrolled in this study (68 years old, 62.7kg (patient 1)and 63 years old, 56.6 kg (patient 2), both are male) were used for thevalidation of the proposed method. The dynamic PET examinatio(uEXPLORER, United Imaging Inc of 1-hour for all patients have beenperformed immediately after an intravenous injection of 18F-FDG with thedose of 0.1 mCi/kg. Such data have been divided into 28 frames (5s × 4, 10s× 4, 30s × 2, 60s × 8 and 300s × 10), and the applied reconstruction methodis 3D TOF list-mode ordered-subsets expectation maximization (OSEM)algorithm combining necessary corrections including normalization, scatter,attenuation and random. Image were reconstructed into a 256 × 256 × 673matrix with 2.53-mm cubic voxels; Parametric analysis were performedbased on the brain regions images while the blood input functions wereextracted from a 4 × 4 × 4 VOI in the thoracic aorta. Totally 19 and 23effective frames, started from the peak value of the blood input functions ofthe two scans, respectively, were included in the analyzed dynamic imageseries. Patlak fitting and iterative calculation of Ki and b were based on thelast 10 frames of the image series. Algorithm Iterative method was used forcalculating the parametric images Ki and b based on a linear controllingequation relating the dynamic pixel radioactive concentrations and theobjective metabolic parameters. Two integral vectors calculated from theblood input function were Kronecker multiplied with unit matrices and thenregarded as the coefficient matrices of the linear controlling equation.Basically, expectation-maximization (EM) algorithm were used for iterativeupdating of the objective parametric images. Comparison Resultantparametric images given by Patlak fitting and the proposed iterative methodwere compared to each other directly. And for the pixels where evidentdifferent Ki values appear, Patlak plots were given for the demonstration ofthe reason for the distinction. Results: Overall speaking, Patlak fitting and the proposed iterative methodgive resultant parametric images that are comparable to each other.However, difference images between Ki images from these methods showsthat there exist pixels that Patlak fitting and iterative method give differentestimated physiological rate values. The comparison of the Ki and b valueson the Patlak plot of these pixels tells that the abnormal data point thatdirectly affect the Patlak fitting result will not seriously hinder our iterativemethod from getting reliable Ki values. Conclusions: The proposed iterative method for parametric imagescalculation from dynamic PET data is a superior method compared to directPatlak fitting since the fortuitous data anomaly in temporal dimension that willsignificantly affect the direct fitting can be eliminated, and more reliableparametric images can be expected.