Acceleration of Radiological Mapping by Image Super Resolution

• Main Authors: Pan, L., Vavrek, J. R.
• Format: Conference Proceeding
• Language: English
• Subjects: Adaptation models; Computational modeling; Radioactive materials; Semiconductor detectors; Semiconductor device measurement; Solid modeling; Three-dimensional displays
• ISSN: 2577-0829
• DOI: 10.1109/NSS/MIC/RTSD57108.2024.10654878

Mapping and estimation of the absolute intensity distribution of radioactive materials in diverse environments see important applications in various fields. A mapping analysis models the environment as a 3-dimensional (3D) grid of voxels, and reconstruct the activity in each voxel based on radiation measurement by a detector that moved across the environment. A high-fidelity reconstruction uses more voxels, which is computationally expensive and difficult for real-time processing. Here, we investigated the application of machine learning method for the acceleration of radiological map reconstruction process. Specifically, instead of reconstructing a high-fidelity map, we first reconstructed a low-fidelity map with fewer number of voxels. Then the image super resolution technique is used to upscale the low-fidelity map to high-fidelity map with more voxels. We adapted and trained a super resolution model with synthetic and real radiological map data. Through qualitative and quantitative evaluation, the super resolution model produced promising performance for the image upscaling. Our approach here can reduce the computational cost significantly, which can enable real-time or near real-time reconstruction of high-fidelity map with lower computer memory and battery usage required.