Fully automated and minimally invasive estimation of the input function and of the regional glucose consumption in [18F]-FDG PET exams of the human brain

Objectives Fully quantitative analysis using PET in the human brain requires the estimation of the pharmacological parameters describing the underlying physiological processes. For this purpose, the arterial input function (IF) can be measured in the wrist artery, but this operation, traumatic and potentially dangerous, is unusable in clinics. In this work, we propose a full process that estimates the IF and the parameters, taking as input a PET image, a T1 MRI image and one venous blood sample.

Methods The materials included 4 healthy subjects having undergone a T1 MRI acquisition and a [18F]-FDG PET acquisition on the ECAT HR+ (Siemens, 4.5mm intrinsic resolution). Arterial and venous blood samples were taken during the acquisition. The method is composed of four parts: (1) registration using a rigid transformation and the mutual information similarity criterion, (2) automated segmentation of the T1 MRI image into 16 brain structures using the SWiPAS method 1, (3) automated extraction of partial volume corrected brain structures TACs using a improvement 2 (GTM20) of the Geometric Transfer Matrix method, and (4) automated estimation of the input function and of the pharmacological parameters using an iterative bootstrap approach 3 (IB-SIME) for the SIME 4 method. Three distinct estimations were performed for each subject.

Results The segmentation was performed successful for all four images. The mean error in area under the input function curve was 4,3%±3,5%, while using GTM instead of GTM20, the IB-SIME method was unable to successfully complete a sufficient number of estimations (n=300), even with a large number of runs.

Conclusions The fully automated extraction of pharmacological parameters is now possible using the proposed process with a mean error lower than 5%. Moreover, this estimation requires only one blood sample, allowing its use in clinics