Event Reconstruction in Radiation Detectors using Convolutional Neural Networks. Banerjee, S., Rodrigues, M., Vija, A. H., & Katsaggelos, A. K. In 2021 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), pages 1–3, oct, 2021. IEEE, IEEE. ![Event Reconstruction in Radiation Detectors using Convolutional Neural Networks [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex Room Temperature Semiconductor Detectors (RTSD) (e.g., CdZnTe and CdZnTeSe) have been recently proposed in novel space, homeland security and medical applications, which provide sub-millimeter position information of interacting $γ$-rays and excellent spectroscopic performance. These detectors have been constructed using a large variety of anode configurations. The virtual Frisch-grid concept with reduced readout channels has been proposed recently. To fully utilize the potential of RTSD, advanced single-polarity charge sensing reconstruction algorithms are needed. Energy and position of interaction reconstruction algorithms rely on physics-based models, with Principal Component Analysis being introduced recently. Proposed deep learning (DL) techniques have the potential to perform event reconstruction with improved position information and better energy resolution than conventional non-DL methods. In this paper, we present a novel DL approach based on Convolutional Neural Networks (CNN) for identifying the energy deposition and position of interaction of the $γ$-rays within the RTSD. The network is trained with input-output data pairs. The input data consists of signals at the electrodes corresponding to each incident event and the output data the position and energy spectrum of those events. Our network consists of 5 stages of convolutional layers, each followed by a batch normalization layer and a max-pooling layer. These layers extract features from the input signals fed to the model. This is followed by 2 stages of fully connected layers. Our model outputs the interaction positions and energies within the RTSD. The model is trained using gradient descent steps using the backpropagation method in Tensorflow library of Python. The network has been tested with unseen signals. The Root Mean Squared Error (RMSE) for test cases were around 1% or less for both position and energy interactions.
@inproceedings{banerjee2021event,
abstract = {Room Temperature Semiconductor Detectors (RTSD) (e.g., CdZnTe and CdZnTeSe) have been recently proposed in novel space, homeland security and medical applications, which provide sub-millimeter position information of interacting $\gamma$-rays and excellent spectroscopic performance. These detectors have been constructed using a large variety of anode configurations. The virtual Frisch-grid concept with reduced readout channels has been proposed recently. To fully utilize the potential of RTSD, advanced single-polarity charge sensing reconstruction algorithms are needed. Energy and position of interaction reconstruction algorithms rely on physics-based models, with Principal Component Analysis being introduced recently. Proposed deep learning (DL) techniques have the potential to perform event reconstruction with improved position information and better energy resolution than conventional non-DL methods. In this paper, we present a novel DL approach based on Convolutional Neural Networks (CNN) for identifying the energy deposition and position of interaction of the $\gamma$-rays within the RTSD. The network is trained with input-output data pairs. The input data consists of signals at the electrodes corresponding to each incident event and the output data the position and energy spectrum of those events. Our network consists of 5 stages of convolutional layers, each followed by a batch normalization layer and a max-pooling layer. These layers extract features from the input signals fed to the model. This is followed by 2 stages of fully connected layers. Our model outputs the interaction positions and energies within the RTSD. The model is trained using gradient descent steps using the backpropagation method in Tensorflow library of Python. The network has been tested with unseen signals. The Root Mean Squared Error (RMSE) for test cases were around 1% or less for both position and energy interactions.},
author = {Banerjee, Srutarshi and Rodrigues, Miesher and Vija, Alexander Hans and Katsaggelos, Aggelos K.},
booktitle = {2021 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC)},
doi = {10.1109/NSS/MIC44867.2021.9875945},
isbn = {978-1-6654-2113-3},
month = {oct},
organization = {IEEE},
pages = {1--3},
publisher = {IEEE},
title = {{Event Reconstruction in Radiation Detectors using Convolutional Neural Networks}},
url = {https://ieeexplore.ieee.org/document/9875945/},
year = {2021}
}
Downloads: 0
{"_id":"J5tZAtA5686XdYZBq","bibbaseid":"banerjee-rodrigues-vija-katsaggelos-eventreconstructioninradiationdetectorsusingconvolutionalneuralnetworks-2021","author_short":["Banerjee, S.","Rodrigues, M.","Vija, A. H.","Katsaggelos, A. K."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"Room Temperature Semiconductor Detectors (RTSD) (e.g., CdZnTe and CdZnTeSe) have been recently proposed in novel space, homeland security and medical applications, which provide sub-millimeter position information of interacting $γ$-rays and excellent spectroscopic performance. These detectors have been constructed using a large variety of anode configurations. The virtual Frisch-grid concept with reduced readout channels has been proposed recently. To fully utilize the potential of RTSD, advanced single-polarity charge sensing reconstruction algorithms are needed. Energy and position of interaction reconstruction algorithms rely on physics-based models, with Principal Component Analysis being introduced recently. Proposed deep learning (DL) techniques have the potential to perform event reconstruction with improved position information and better energy resolution than conventional non-DL methods. In this paper, we present a novel DL approach based on Convolutional Neural Networks (CNN) for identifying the energy deposition and position of interaction of the $γ$-rays within the RTSD. The network is trained with input-output data pairs. The input data consists of signals at the electrodes corresponding to each incident event and the output data the position and energy spectrum of those events. Our network consists of 5 stages of convolutional layers, each followed by a batch normalization layer and a max-pooling layer. These layers extract features from the input signals fed to the model. This is followed by 2 stages of fully connected layers. Our model outputs the interaction positions and energies within the RTSD. The model is trained using gradient descent steps using the backpropagation method in Tensorflow library of Python. The network has been tested with unseen signals. The Root Mean Squared Error (RMSE) for test cases were around 1% or less for both position and energy interactions.","author":[{"propositions":[],"lastnames":["Banerjee"],"firstnames":["Srutarshi"],"suffixes":[]},{"propositions":[],"lastnames":["Rodrigues"],"firstnames":["Miesher"],"suffixes":[]},{"propositions":[],"lastnames":["Vija"],"firstnames":["Alexander","Hans"],"suffixes":[]},{"propositions":[],"lastnames":["Katsaggelos"],"firstnames":["Aggelos","K."],"suffixes":[]}],"booktitle":"2021 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC)","doi":"10.1109/NSS/MIC44867.2021.9875945","isbn":"978-1-6654-2113-3","month":"oct","organization":"IEEE","pages":"1–3","publisher":"IEEE","title":"Event Reconstruction in Radiation Detectors using Convolutional Neural Networks","url":"https://ieeexplore.ieee.org/document/9875945/","year":"2021","bibtex":"@inproceedings{banerjee2021event,\nabstract = {Room Temperature Semiconductor Detectors (RTSD) (e.g., CdZnTe and CdZnTeSe) have been recently proposed in novel space, homeland security and medical applications, which provide sub-millimeter position information of interacting $\\gamma$-rays and excellent spectroscopic performance. These detectors have been constructed using a large variety of anode configurations. The virtual Frisch-grid concept with reduced readout channels has been proposed recently. To fully utilize the potential of RTSD, advanced single-polarity charge sensing reconstruction algorithms are needed. Energy and position of interaction reconstruction algorithms rely on physics-based models, with Principal Component Analysis being introduced recently. Proposed deep learning (DL) techniques have the potential to perform event reconstruction with improved position information and better energy resolution than conventional non-DL methods. In this paper, we present a novel DL approach based on Convolutional Neural Networks (CNN) for identifying the energy deposition and position of interaction of the $\\gamma$-rays within the RTSD. The network is trained with input-output data pairs. The input data consists of signals at the electrodes corresponding to each incident event and the output data the position and energy spectrum of those events. Our network consists of 5 stages of convolutional layers, each followed by a batch normalization layer and a max-pooling layer. These layers extract features from the input signals fed to the model. This is followed by 2 stages of fully connected layers. Our model outputs the interaction positions and energies within the RTSD. The model is trained using gradient descent steps using the backpropagation method in Tensorflow library of Python. The network has been tested with unseen signals. The Root Mean Squared Error (RMSE) for test cases were around 1% or less for both position and energy interactions.},\nauthor = {Banerjee, Srutarshi and Rodrigues, Miesher and Vija, Alexander Hans and Katsaggelos, Aggelos K.},\nbooktitle = {2021 IEEE Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC)},\ndoi = {10.1109/NSS/MIC44867.2021.9875945},\nisbn = {978-1-6654-2113-3},\nmonth = {oct},\norganization = {IEEE},\npages = {1--3},\npublisher = {IEEE},\ntitle = {{Event Reconstruction in Radiation Detectors using Convolutional Neural Networks}},\nurl = {https://ieeexplore.ieee.org/document/9875945/},\nyear = {2021}\n}\n","author_short":["Banerjee, S.","Rodrigues, M.","Vija, A. H.","Katsaggelos, A. K."],"key":"banerjee2021event","id":"banerjee2021event","bibbaseid":"banerjee-rodrigues-vija-katsaggelos-eventreconstructioninradiationdetectorsusingconvolutionalneuralnetworks-2021","role":"author","urls":{"Paper":"https://ieeexplore.ieee.org/document/9875945/"},"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://sites.northwestern.edu/ivpl/files/2023/06/IVPL_Updated_publications-1.bib","dataSources":["ePKPjG8C6yvpk4mEK","37Qfzv6wRptYkoSCL","zFPgsTDAW8aDnb5iN","E6Bth2QB5BYjBMZE7","nbnEjsN7MJhurAK9x","PNQZj6FjzoxxJk4Yi","7FpDWDGJ4KgpDiGfB","bod9ms4MQJHuJgPpp","QR9t5P2cLdJuzhfzK","D8k2SxfC5dKNRFgro","7Dwzbxq93HWrJEhT6","qhF8zxmGcJfvtdeAg","fvDEHD49E2ZRwE3fb","H7crv8NWhZup4d4by","DHqokWsryttGh7pJE","vRJd4wNg9HpoZSMHD","sYxQ6pxFgA59JRhxi","w2WahSbYrbcCKBDsC","XasdXLL99y5rygCmq","3gkSihZQRfAD2KBo3","t5XMbyZbtPBo4wBGS","bEpHM2CtrwW2qE8FP","teJzFLHexaz5AQW5z"],"keywords":[],"search_terms":["event","reconstruction","radiation","detectors","using","convolutional","neural","networks","banerjee","rodrigues","vija","katsaggelos"],"title":"Event Reconstruction in Radiation Detectors using Convolutional Neural Networks","year":2021}