A bootstrap method for a totally non-invasive input function and pharmacokinetic parameters estimation in 18F-FDG PET images of the human brain

The pharmacokinetics extracted from PET images in the brain is relied to the physiological mechanisms that describe the processing of the tracer by the brain structures. These processes can be described through a multi-compartments model detailing the different states of the tracer and the transfer rate between these states. Numerous work have been proposed for the estimation of the input function (IF) from the internal carotids in the brain, but these methods are not able to extract efficiently the individual parameters of the model. A method has been proposed for the simultaneous estimation of the input function (IF) and of the pharmacokinetic parameters (PK) but since it is highly sensitive to the noise that affects the input time activity curves (TACs), it has to resort to venous blood sampling. This paper proposes an improvement of the SIME method avoiding the resort to venous blood samples using a bootstrap method. The SIME method, although of great interest for the input function estimation from PET images, suffers from an important bias when the structures TACs are affected with noise or spillover. The bootstrap approach proposed in this work allowed an important reduction of the error committed on the estimation of the input function under realistic spillover and noise. The results obtained on the simulated TACs datasets suggested the possible use of this Bootstrap-SIME method without resort to blood samples. The present results show that the input TACs have to be corrected for spillover prior to any IF estimation without resort to blood samples. The method precision will also be assessed on experimental 18F-FDG PET images in the human brain and in the rat whole-body segmented using automated methods. On the real PET exams however, one blood sample was necessary for the estimation of a correct input function. The estimation was correct for three subjects over four and for the fourth, the estimation was correct after 15 minutes. This approach may also be use- - d in conjunction with more blood samples, which would stabilize the elimination part of the IF curve.