Journal of Molecular Imaging & Dynamics

Journal of Molecular Imaging & Dynamics
Open Access

ISSN: 2155-9937


Non-Negative Matrix Factorization Based Input Function Extraction for Mouse Imaging in Small Animal PET - Comparison with Arterial Blood Sampling and Factor Analysis

Dominik Schulz, Arne Tapfer, Andreas Buck, Sybille Reeder, Matthias Miederer, Eliane Weidl, Sibylle I. Ziegler, Markus Schwaiger and Ralph A. Bundschuh

Objectives: Retrieving the accurate time-tracer activity concentration curve of the blood (arterial input function) is mandatory for performing bio kinetic model analysis of dynamic PET data. Especially in small rodents, gathering the input function remains an active area of research as no generally applicable solution was found so far. While surgically catheterizing blood vessels of rodents is possible, it is labour intensive and time resolution of blood sampling is restricted due to the limited amount of overall blood and the procedure of blood withdrawal itself. Obtaining the input function from the PET images themselves seems thus to be favourable, but suffers from several factors, one of them being spill-in of adjacent tissues. Particularly in mice and for [18F]FDG, the spill-in complicates using the time-activity curve (TAC) from a region of interest (ROI) over the left ventricle (LV) because the signal of the ROI contains contributions from both, myocardial uptake as well as arterial blood activity. We propose non-negative matrix factorization (NMF) as an image based algorithm for separating myocardial tracer concentration from the blood input function. The aim of this study was to evaluate the potential of NMF as an image based algorithm for retrieving the input function by comparison with blood sampling and Factor Analysis (FA).

Method: The femoral arteries of eight mice were surgically catheterized. With the injection of [18F]FDG, a 60 minute PET scan was started during which arterial blood samples were manually drawn from the catheter. For analysis, NMF and FA were performed in a ROI placed over the LV. The NMF algorithm shares similarities with principal component analysis and FA, the advantage over the later two being its non-negativity constraint. For normalization of the NMF extracted curve, the peak value of tracer activity in an early image of the LV and a late blood sample was used. The normalized NMF curve was visually compared to the TAC retrieved from the blood samples and to the FA retrieved TAC. For a quantitative comparison of performance, Pearson correlation and square-root of sum of squares (RSS) between NMF/FA and blood sampling curves was calculated.

Results: TAC based on NMF, FA and arterial blood samples were obtained and compared in all 8 mice. The NMF derived curves described the blood sampling based curves visually significantly better than FA. Pearson correlation between NMF and blood sampling curves ranged from 0.21 to 0.92 with an average of 0.69. Pearson correlation for FA ranged from 0.46 to 0.81 with an average of 0.65. Mean RSS was 2.70E + 006 for NMF and 3.40E + 006 for FA.

Conclusion: In the examined parameters, visual accordance, Pearson correlation and RSS, NMF performs superior to FA and seems to be a promising method for the extraction of the input function from PET images of small rodents without the need for arterial blood sampling.