|
Photon counting detectors (PCDs) offer significant advantages in quantitative material decomposition applications due to their spectroscopic capabilities. However, various challenges, including charge sharing, partial charge collection, fluorescence photons from the detector’s material, and crystal defects introduced during fabrication, negatively impact image quality and energy resolution. Consequently, conventional flat-field correction proves insufficient for accurate spectral correction. A common correction technique is the signal-to-thickness calibration (STC) method, which correlates photon counts with material thickness. In biological applications, polymethyl methacrylate (PMMA) is often used as a calibration material due to its similar attenuation properties to soft tissue. However, single-material calibration struggles to yield accurate results for samples with higher atomic numbers, such as bone. In this study, we present a deep neural network model trained using two calibration materials (PMMA and Aluminum) that not only corrects the measured mass attenuation of complex materials but also restores photon counts, leading to improved accuracy in energy spectral measurements. When tested using the Medipix3 CdTe photon counting detector, this method successfully removed the intrinsic x-ray fluorescence of CdTe. Additionally, our approach offers the advantage of not requiring a detailed formulation of the detector’s response function. Our experimental results demonstrate that this approach reduces noise and enhances mass attenuation accuracy compared to single-material calibration. By incorporating both calibration datasets during training, we developed a single model capable of handling a wide range of potential attenuation values. The dual-material model method addresses the limitations of conventional techniques and holds promise for improving the performance of photon-counting detectors across various applications.
|
