ParaPET, a new methodology to derive parametric PET images of FDG kinetics based on a mathematical approach integrating an error model of measurements

Objectives: We propose a method derived from Barbolosi's approach [1,2] to generate parametric PET images of FDG kinetics: ParaPET. Our primary objective was to compare our methodology with Patlak [3] and Hunter [4] analyses in the evaluation of non-small cell lung cancers (NSCLC). Methods: ParaPET is based on a mathematical approach integrating an error model of PET measurements and a non-invasive image derived blood activity (IDBA) from dynamic PET images of the aorta leading to measure kinetic parameters at the voxel level. ParaPET approach was evaluated along with Hunter and Barbolosi (with blood samples or with IDBA) methods, using Patlak analysis as a reference. These methods were implemented on a Java based plug-in. Two FDG PET/CT acquisitions are required for validation purpose. First, an early 30-min list-mode PET/CT acquisition centered over the heart is done starting simultaneously with the FDG injection. This first acquisition allows the determination of the early IDBA required for Patlak analysis. Second, a 15-min list-mode PET/CT acquisition, centered over the lesions, is done 80 min after injection. The late acquisition is associated with five venous blood samples (each 3 min) to implement Patlak, Hunter and Barbolosi methods. The late IDBA is also used to determine the mean FDG activity concentration in blood and its variability needed for the error model in ParaPET. The ParaPET approach generates a set of parametric images of FDG kinetics for the entire lesion. Two kinetic parameters were investigated: Ki, the net influx rate constant and Kimax corresponding to the maximum value of Ki in the lesion. Results: The methods were evaluated on three NSCLC patients planned for radiochemotherapy included in the on-going ParaPET clinical trial (NCT 02821936). Seven tumour lesions were evaluated on PET images. Except Hunter's method which presented a correlation coefficient of 0.805 with Ki value from Patlak, the others quantification approaches correlated well with Patlak (r2 = 0.968 and r2 = 0.989 respectively for Barbolosi and Barbolosi w/ IDBA) but our approach showed the best correlation with the reference (r2 = 0.998). The mean errors (± SD) in Kimax estimates were 31.5% ± 21.1% for Hunter, -7.0% ± 10.0% for Barbolosi, 4.5% ± 5.0% for modified Barbolosi (w/ IDBA) and 3.4% ± 3.6% for ParaPET. These results confirmed that dynamic PET images over the heart is a solution to avoid invasive blood samples. Conclusion: ParaPET approach, compared with