Simulation of PET data

Simulation of dynamic PET data is an essential part in developing and validation of analysis models and software. General idea is that with a flawless input data (plasma and/or blood curves), comprehensive compartmental model, and certain set of model parameters, a flawless tissue curve can be calculated. Different analysis methods can then be tested with the simulated tissue data, trying to reproduce the applied model parameters, or at least the physiologically most relevant parameter. Measurement noise can be added to simulated tissue and input data to study noise-induced biases and to compare the noise-sensitivity of analysis methods.

As a rule the steps are:

1. Creating representative input data
2. Determining the appropriate model and model parameters
3. Simulation of tissue curves using compartmental model
4. Simulation of PET time frames
5. Simulation of measurement noise
6. Analysis of created datasets with the methods to be tested

Dynamic PET images can be simulated as well. Regions of interest in PET image can be filled with the tissue curves from step 3, and noise can then be added to the dynamic image.

The performance of PET scanners and reconstruction algorithms can be simulated by using mathematical phantoms of the brain and heart, where predefined regions are filled with flawless concentration curves, and which then can calculate dynamic PET sinograms or images for further analysis.