X-ray spectra correction based on deep learning CNN-LSTM model

Ma, Xing-Ke; Hong-quan, Huang; Huang, Bo-Rui; Shen, Zhi-Wen; Wang, Qingtai; Xiao, Yao; Zhong, Cheng-Lin; Xin, Hao; Sun, Peng; Kai-ming, Jiang; Tang, Lin; Ding, Wei-Cheng; Zhou, Wei; Zhou, Jianbin

doi:10.1016/j.measurement.2022.111510

Cited by 5 publications

(3 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More recently, Holbrook et al [181] used multienergy CT scans with an EID to calibrate the PCD spectral distortion, and adopted a U-Net to map the distorted PCD projections into monochromatic projections generated by multienergy CT scans after material decomposition. Ma et al [182] introduced CNN-LSTM to correct pulse pileup distortion in Xray source spectrum measurements, while Smith et al anti-scatter grids relative to the small detector pixel size, resulting in improved image quality and HU value accuracy. Due to the complexity of PCDs, their pixels tend to suffer more nonuniformity due to detector imperfections compared to EIDs, making the ring artifact issues more prominent in PCCT.…”

Section: A Pcct Data Pre-processingmentioning

confidence: 99%

Systematic Review on Learning-Based Spectral CT

Bousse,

Kandarpa,

Rit

et al. 2024

IEEE Trans. Radiat. Plasma Med. Sci.

View full text Add to dashboard Cite

Spectral computed tomography (CT) has recently emerged as an advanced version of medical CT and significantly improves conventional (single-energy) CT. Spectral CT has two main forms: dual-energy computed tomography (DECT) and photon-counting computed tomography (PCCT), which offer image improvement, material decomposition, and feature quantification relative to conventional CT. However, the inherent challenges of spectral CT, evidenced by data and image artifacts, remain a bottleneck for clinical applications. To address these problems, machine learning techniques have been widely applied to spectral CT. In this review, we present the state-of-the-art data-driven techniques for spectral CT.

show abstract

Section: A Pcct Data Pre-processingmentioning

confidence: 99%

Systematic Review on Learning-Based Spectral CT

Bousse,

Kandarpa,

Rit

et al. 2024

IEEE Trans. Radiat. Plasma Med. Sci.

View full text Add to dashboard Cite

show abstract

“…where 𝑘 is the total number of pulses, ℎ 𝑡 is the actual height of the 𝑖 th pulse, and ℎ 𝑝 is the predicted height of the 𝑖 th pulse. 16)_LP (12,16,20,24,28,32,36) Optimize input sliced pulse data…”

Section: Data Preprocessing For Establishing Training and Test Datasetmentioning

confidence: 99%

“…In recent years, deep learning technologies have been applied to various fields of radiation measurement, such as radiation imaging, dosimetry, spectral data analysis, and pulse signal analysis, owing to their recognized characteristics, including feature extraction of complex patterns, nonlinear responses, handling of large-scale data, high estimation accuracy, and the ability to solve various problems [13][14][15][16][17][18]. Among them, the deep learning-based PHE method presented by Jeon et al, (2022) has demonstrated high correction performance and can be universally applied without limitations to specific detectors [17,18].…”

Section: Introductionmentioning

confidence: 99%

Optimizing deep learning-based piled-up pulse height correction method for high radiation-field application

Kim,

Ko,

Lee

et al. 2023

J. Inst.

View full text Add to dashboard Cite

In high-radiation environments, measured pile-up pulses can lead to unavoidable issues such as total count loss and spectrum distortion. Additionally, the recording of large volumes of data within a short period makes real-time processing difficult. In this study, a deep learning-based pulse height estimation (PHE) method was optimized to perform pile-up signal correction in high-radiation fields. First, we adopted a previous deep-learning-based PHE method that allows for fast correction without being restricted to specific detectors. However, the peak-finding method was slightly modified to improve the count restoration rate. Moreover, the input data length of the deep learning model was optimized for convolutional neural networks (CNN) and deep neural networks (DNN) to achieve the maximum correction performance using minimal input data. A series of single pulses was experimentally obtained from a LaBr3 detector with a short decay time to prepare a dataset for training the deep learning models. The pile-up signals were generated by randomly synthesizing single pulses. Samples around their peaks were sliced using the peak-finding method and used as input data for the deep learning models. As a result of the optimization, the modified peak-finding method improved the count restoration rate compared to the previous method by effectively detecting the peaks of tail pile-up, and peak pile-up pulses. Furthermore, the input data length and region were optimized based on the performance evaluation of each deep learning model. Despite having a simpler architecture than the CNN model, the DNN model demonstrated excellent PHE performance. The results of this study showed the efficient and practical considerations necessary for applying pile-up signal correction in high-radiation fields.

show abstract