Charge sharing energy correction in CdTe sensors using machine learning and its application in SPECT imaging

Small-pixel CdTe/CZT detectors based multi-Isotope hyperspectral SPECT imaging systems suffering serious degradation of energy performance and detection efficiency due to the incomplete charge collection. The charge sharing events collected by more than one pixel usually lose energy which is relevant to the collected energy ratio. When the interaction of photon is closer to the metal contact gap between metal contacts and loss more energy, the collected energy ratio by several pixels is more evenly distributed. In this thesis, a multi-pixel charge sharing energy reconstruction method based on Fully Connected Neural Network (FCNN) was developed and can flexibly reconstruct multi-pixel charge sharing events energy. The proposed FCNN charge sharing correction algorithm could enhance the spectral performance of the small-pixel CdTe semiconductor detectors by reconstructing bi-pixel, triple-pixel, and quad-pixel charge sharing events. Compared to the traditional charge-sharing discrimination (CSD) method, the correction of the charge-sharing events could increase the sensitivity of SPECT system with higher detection efficiency. In this study, we will compare the pure energy reconstruction and combination reconstruction energy reconstruction method implying bi-pixel, tri-pixel, and quad-pixel events and the results of traditional charge sharing addition (CSA) and charge sharing discrimination (CSD) methods. The machine learning method shows significant flexibility in multi-pixel charge sharing energy reconstruction and potential in sub-pixel SPECT imaging. Our future work is planned to investigate the application of FCNN on position estimation for increasing of the spatial resolution of the SPECT system.U of I OnlyAuthor requested U of Illinois access only (OA after 2yrs) in Vireo ETD syste