A Systematic Evaluation of 18F Positron Imaging Using Radioluminescence Microscopy by Computational Simulation

Objectives Radioluminescence microscopy, an emerging live-cell imaging modality, can visualize the distribution of beta-emitting radiotracers with high resolution. It has been used for monitoring cancer metabolism and drug binding on a single cell level. However, our theoretical understanding of how the radioluminescence signal is generated remains incomplete. In order to optimize the experimental setup and the processing of raw camera images, a computational simulation of 18F positron imaging using radioluminescence microscopy is carried on.

Methods The simulation comprises three consecutive modules. First, the transport of the ionizing radiation through the cell and the scintillator is performed using the GEANT4 Monte-Carlo simulation package. The resulting ionization is used to compute the production of scintillation light. The propagation of this scintillation light through the microscope is modeled by a convolution with a point spread function (generated by Gibson and Lanni model), which models the response of the microscope. Finally, the physical effect of measuring the scintillation light using an electron-multiplicative charge-coupled device (EMCCD) camera is modeled using a stochastic numerical photosensor model (multiple noises and electron multiplication are considered). The simulated output of the EMCCD camera is further processed using the ORBIT image reconstruction methodology to evaluate the endpoint images.

Results Based on simulations, a spatial resolution of 18.5 µm can be achieved, including image reconstruction (which improves the spatial resolution by 15 µm on average). This compares well with the experimentally measured spatial resolution of 22 µm. The system sensitivity is estimated to be 21% under realistic simulation parameters and is consistent with our previous experiments. Furthermore, the sensitivity decreases by 1% for every 1 µm the radioactive source is moved away from the scintillator. The properties of individual ionization tracks were also quantified. The mean area and intensity of all pixels within a positron track are 77.2 square µm and 13300 counts/incidence, respectively in the simulation, which is 8.7% higher and 9.3% lower than the experimental values. The mean and standard deviation of simulated dark noises are 3048 counts/pixel and 40.0 counts/pixel using maximum EMCCD gain, which is consistent with the properties of the camera used in experiments. Finally, an image of a sparse set of single cells is simulated and is visually similar to the measured cell image. Moreover, the energy limitation of detectable positron (around 8 keV), the positron penetration depth (mean=21.5 µm, maximum=277.8 µm), the influences of system factors (such as noises) on spatial resolution and etc. are also estimated.

Conclusions Our simulation methodology agrees with experimental measurements taken with radioluminescence microscopy. This in silico approach can be used to guide further instrumentation developments, and it also provides a framework for improving image reconstruction. This will help improve the quality of single cell imaging and facilitate a deeper understanding of tumor cell biology.