Modeling and Optimization of a Gamma Camera and a Novel Multi-Pinhole Collimator for Localization of Sentinel Lymph Nodes

Detector systems and subsequent image formation methods are important in many applications today including medical imaging and high energy physics experiments. These detectors can be designed to satisfy novel tasks and applications; they are often modeled using Monte Carlo methods and can then be optimized using machine learning. In this work, we first use Geant4 to model a Dilon 6800 gamma camera detector with a novel multi-pinhole collimator (31 pinholes in a 26x19 cm area). We then use this model to mock a typical clinical situation of single and multiple sentinel lymph nodes during SLN biopsy. The data from this simulation is used to optimize reconstruction of test sources in a set volume of interest using a machine learning algorithm. With traditional stationary operation of the system, location of the reconstructed one point-source agrees well with the known position of the source. Agreement of reconstruction of two point-sources with known positions, to mimic a more realistic clinical situation, is also reported. In conclusion, our Geant4 model and the associated convolutional neural network reconstruct accurate models of point sources in a volume of interest from a simulated projection image.