Proton imaging with machine learning

The purpose of this study is to introduce a compact calorimeter that can offer an additional imaging tool for proton therapy centers. The tungsten, gadolinium, and lanthanide based high-density scintillating glass designed for this purpose has the ability to stop 200 MeV protons with thicknesses less than 60 mm, which allows us to model a compact detector that can be attached to a gantry. The details of the glass development and preliminary imaging efforts with this detector were previously reported. This study summarizes the Artificial Neural Network based imaging efforts with this novel proton imager detector. A library of proton conical beam CT (CBCT) scans of 800 tumors was created via GATE simulations. This tumor library was used for training purposes with two different machine learning tools, Flux and PyTorch. Here, the proof-of-concept machine learning imaging study is reported. The novel material development, compact detector design, and machine learning based imaging can make this approach useful for clinical applications.