Improvement of coded aperture imaging with 3D U-NET and partitioned G(E) function in high background environment

Abstract

radiation doses for workers and significantly degrade the imaging performance of radiation imaging equipment. Firstly, the unknown location of radioactive sources leads to inaccurate dose estimations based on the shape and direction of the detector using conventional G(E) function. Therefore, this study proposes a method that utilizes multiple G(E) functions to achieve more accurate dose estimations. The investigation demonstrates that the proposed partitioned G(E) functions improve dose estimation accuracy by more than 1.5 times compared to conventional G(E) function when the distribution of radioactive sources is unknown. Furthermore, while conventional G(E) functions produce large errors in specific directions or energy ranges, the proposed partitioned G(E) functions provide dose estimations with uniform errors. Additionally, signals outside the Field of View (FOV) create significant artifacts in radiation imaging, making it difficult to estimate hotspots. To address this, we employs 3D-Unet to separate and reconstruct clear hotspots by extracting coded signals without the need for additional structures. Although the proposed method exhibits slightly lower performance compared to the Anti-mask method at high-energy gamma rays, it possesses the advantage of avoiding false hotspot estimations.