Deep Image Prior (PyTomography Example)

The following repository contains all necessary files to perform the DIPRecon algorithm in PyTomography. The phantom corresponds to an ultra high resolution PET/MRI brain scan. PET simulation was performed via GATE.

The files are ran in the following order:

• compute_eta.py computes the normalization factor $\eta$ needed for PET reconstruction. It uses data from a cylindrical calibration scan performed using the same system geometry.
• open_data.py loads and processes the raw ROOT coincidence data and converts it into a readable .npy file.
• init_recon.py performs an initial reconstruction of the data using OSEM
• dip_init_net.py trains a DIP network to predict the output of the init_recon.py file using an MRI image as input. This is used to get an initial network configuration before running the next script.
• dip_recon.py performs DIP reconstruction.

Other files include:

• simulation_parameters.py controls all parameters used in the code.
• misc.py contains all miscellaneous functions used for image alignment/etc required
• networks.py contains the neural network architecture used here.
• analysis.ipynb contains code for analyzing the reconstructions/statistics.