Derived input function from dynamic cardiac 18F-FDG PET images in rodents based on the generalized Gaussian ICA model

The aim of this study is to develop a new method to extract input function (IF) from dynamic positron emission tomography (PET) data using the well known [¹⁸F]-2-Deoxy-2-fluoro-d-glucose (¹⁸F-FDG) for the determination of the myocardial metabolic rate of glucose (MMRG). In the case of cardiac studies, the IF can be obtained directly from the data by mean of a region of interest (ROI) drawn over the blood pool. However, in small animal imaging, hearts and arteries are small compared to the scanner spatial resolution. Consequently, vascular radioactivity is spilled over adjacent tissues and vice versa. As a result, curves obtained from ROIs are made up of a mixture of clean blood activity and the surrounding tissues activities. In this work, we assume that the elementary activities of the ventricular blood pool and the myocardium are spatially independent and that dynamic PET image frames are composed of a mixture of them. We use the independent component analysis (ICA) and a Generalized Gaussian Distribution (GGD) model in order to reveal the unknown de-mixing matrix. The derived ICA-IF are then compared to IF derived from invasive arterial blood samples. Results display similar shapes to those obtained by arterial sampling. Moreover, the MMRG calculated using the ICA-IF and two-compartment model correlated well with the results obtained from the sampled IF (2% difference only).