Purpose: Conventional electronic portal imaging devices (EPIDs) use a thin layer (<1mm) of scintillator to convert x-rays to visible light that is then detected by an array of photodetectors. Consequently, current EPIDs suffer poor contrast-to-noise ratios because of their limited (1-2%) detective quantum efficiency (DQE) and offer no spectral information. Spectral imaging in megavoltage imaging has been limited to kV/MV switching or expensive, non-continuous, multi-layered scintillator detectors. We are seeking an order-of-magnitude increase in the x-ray detection layer to 10-30mm using a transparent scintillator, capturing the emitted 4D light field using a prosumer light-field (LF) camera, and performing computational refocusing to assess the spectral information throughout the scintillator volume. The purpose of this work was to develop and validate a forward model of the proposed imaging system.
Methods: A Lytro Illum LF camera was used to image a 22mm-thick LKH-5 scintillator. A Varian LINAC and small animal irradiator, XRAD225Cx, were used to generate 6-MV/225-kV x-ray beams, respectively. The LF forward model utilized open-source Python packages for ray-tracing and lens modeling. Scintillation light was analytically modeled. 3D printed phantoms were used to validate the forward model versus experimental observations.
Results: At 225 kV, the Illum was able to perform minor spectral separation throughout the scintillator with a min-max difference of light intensity of ~5%, thus showing the Illum is an unoptimized LF camera for spectral extraction due to the small angular sampling by the microlenses. The Illum was able to be forward modeled giving validity to this LF model approach to simulate a more optimal camera for MV imaging applications.
Conclusion: While applying additional processes such as deconvolution to further improve spectral performance helps, using a single LF camera to extract spectral information from a continuous light-source like a scintillator requires an optimal camera or an ideal image scene for performance.
Funding Support, Disclosures, and Conflict of Interest: Funding for this work was provided by the NIH S10 Award OD025081-01, T32 EB002103, and R21EB028103. Conflict of interest: Patrick La Riviere receives research funding from Accuray, Inc (unrelated to this research) and is a consultant for and has stock options in MetriTrack, Inc. (unrelated to this research).