Pub Date : 2024-06-24DOI: 10.1109/TRPMS.2024.3417901
Lei Bi;Xiaohang Fu;Qiufang Liu;Shaoli Song;David Dagan Feng;Michael Fulham;Jinman Kim
Background: Automated segmentation of multimodality positron emission tomography—computed tomography (PET-CT) data is a major challenge in the development of computer-aided diagnosis systems (CADs). In this context, convolutional neural network (CNN)-based methods are considered as the state-of-the-art. These CNN-based methods, however, have difficulty in co-learning the complementary PET-CT image features and in learning the global context when focusing solely on local patterns. Methods: We propose a cascaded CNN-transformer network (CCNN-TN) tailored for PET-CT image segmentation. We employed a transformer network (TN) because of its ability to establish global context via self-attention and embedding image patches. We extended the TN definition by cascading multiple TNs and CNNs to learn the global and local contexts. We also introduced a hyper fusion branch that iteratively fuses the separately extracted complementary image features. We evaluated our approach, when compared to current state-of-the-art CNN methods, on three datasets: two nonsmall cell lung cancer (NSCLC) and one soft tissue sarcoma (STS). Results: Our CCNN-TN method achieved a dice similarity coefficient (DSC) score of 72.25% (NSCLC), 67.11% (NSCLC), and 66.36% (STS) for segmentation of tumors. Compared to other methods the DSC was higher for our CCNN-TN by 4.5%, 1.31%, and 3.44%. Conclusion: Our experimental results demonstrate that CCNN-TN, when compared to the existing methods, achieved more generalizable results across different datasets and has consistent performance across various image fusion strategies and network backbones.
{"title":"Co-Learning Multimodality PET-CT Features via a Cascaded CNN-Transformer Network","authors":"Lei Bi;Xiaohang Fu;Qiufang Liu;Shaoli Song;David Dagan Feng;Michael Fulham;Jinman Kim","doi":"10.1109/TRPMS.2024.3417901","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3417901","url":null,"abstract":"<italic>Background:</i>\u0000 Automated segmentation of multimodality positron emission tomography—computed tomography (PET-CT) data is a major challenge in the development of computer-aided diagnosis systems (CADs). In this context, convolutional neural network (CNN)-based methods are considered as the state-of-the-art. These CNN-based methods, however, have difficulty in co-learning the complementary PET-CT image features and in learning the global context when focusing solely on local patterns. \u0000<italic>Methods:</i>\u0000 We propose a cascaded CNN-transformer network (CCNN-TN) tailored for PET-CT image segmentation. We employed a transformer network (TN) because of its ability to establish global context via self-attention and embedding image patches. We extended the TN definition by cascading multiple TNs and CNNs to learn the global and local contexts. We also introduced a hyper fusion branch that iteratively fuses the separately extracted complementary image features. We evaluated our approach, when compared to current state-of-the-art CNN methods, on three datasets: two nonsmall cell lung cancer (NSCLC) and one soft tissue sarcoma (STS). \u0000<italic>Results:</i>\u0000 Our CCNN-TN method achieved a dice similarity coefficient (DSC) score of 72.25% (NSCLC), 67.11% (NSCLC), and 66.36% (STS) for segmentation of tumors. Compared to other methods the DSC was higher for our CCNN-TN by 4.5%, 1.31%, and 3.44%. \u0000<italic>Conclusion:</i>\u0000 Our experimental results demonstrate that CCNN-TN, when compared to the existing methods, achieved more generalizable results across different datasets and has consistent performance across various image fusion strategies and network backbones.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"814-825"},"PeriodicalIF":4.6,"publicationDate":"2024-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143751","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-14DOI: 10.1109/TRPMS.2024.3414578
A. Mariscal-Castilla;S. Gómez;R. Manera;J. M. Fernández-Tenllado;J. Mauricio;N. Kratochwil;J. Alozy;M. Piller;S. Portero;A. Sanuy;D. Guberman;J. J. Silva;E. Auffray;R. Ballabriga;G. Ariño-Estrada;M. Campbell;D. Gascón
Time of Flight positron emission tomography (TOF-PET) scanners demand electronics that are power-efficient, low-noise, cost-effective, and possess a large bandwidth. Recent developments have demonstrated sub-100 ps time resolution with elevated power consumption per channel, rendering this unfeasible to build a scanner. In this work, we evaluate the performance for the TOF-PET of the FastIC front-end using different scintillators and silicon photomultipliers (SiPMs). FastIC is an eight-channel application specific integrated circuit developed in CMOS 65 nm capable of measuring the energy and the arrival time of a detected pulse with 12 mW per channel. Using Hamamatsu SiPMs (S13360-3050PE) coupled to LSO:Ce:0.2%Ca crystals of $2times 2times $ 3 mm 3 and LYSO:Ce:0.2%Ca of $3.13times 3.13times $ 20 mm 3 , we measured a coincidence time resolution (CTR) of ( $95~pm ~3$ ) and $156~pm ~4$ ) ps full width half maximum (FWHM), respectively. With Fondazione Bruno Kessler NUV-HD LF2 M0 SiPMs coupled to the same crystals, we obtained a CTR of ( $76~pm ~2$ ) and ( $127~pm ~3$ ) ps FWHM. We employed FastIC with a TlCl pure Cherenkov emitter, demonstrating time resolutions comparable to those achieved with the high-power-consuming electronics. These findings shows that the FastIC represents a cost-effective alternative that can significantly enhance the time resolution of the current TOF-PET systems while maintaining low power consumption.
{"title":"Toward Sub-100 ps TOF-PET Systems Employing the FastIC ASIC With Analog SiPMs","authors":"A. Mariscal-Castilla;S. Gómez;R. Manera;J. M. Fernández-Tenllado;J. Mauricio;N. Kratochwil;J. Alozy;M. Piller;S. Portero;A. Sanuy;D. Guberman;J. J. Silva;E. Auffray;R. Ballabriga;G. Ariño-Estrada;M. Campbell;D. Gascón","doi":"10.1109/TRPMS.2024.3414578","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3414578","url":null,"abstract":"Time of Flight positron emission tomography (TOF-PET) scanners demand electronics that are power-efficient, low-noise, cost-effective, and possess a large bandwidth. Recent developments have demonstrated sub-100 ps time resolution with elevated power consumption per channel, rendering this unfeasible to build a scanner. In this work, we evaluate the performance for the TOF-PET of the FastIC front-end using different scintillators and silicon photomultipliers (SiPMs). FastIC is an eight-channel application specific integrated circuit developed in CMOS 65 nm capable of measuring the energy and the arrival time of a detected pulse with 12 mW per channel. Using Hamamatsu SiPMs (S13360-3050PE) coupled to LSO:Ce:0.2%Ca crystals of \u0000<inline-formula> <tex-math>$2times 2times $ </tex-math></inline-formula>\u0000 3 mm\u0000<sup>3</sup>\u0000 and LYSO:Ce:0.2%Ca of \u0000<inline-formula> <tex-math>$3.13times 3.13times $ </tex-math></inline-formula>\u0000 20 mm\u0000<sup>3</sup>\u0000, we measured a coincidence time resolution (CTR) of (\u0000<inline-formula> <tex-math>$95~pm ~3$ </tex-math></inline-formula>\u0000) and \u0000<inline-formula> <tex-math>$156~pm ~4$ </tex-math></inline-formula>\u0000) ps full width half maximum (FWHM), respectively. With Fondazione Bruno Kessler NUV-HD LF2 M0 SiPMs coupled to the same crystals, we obtained a CTR of (\u0000<inline-formula> <tex-math>$76~pm ~2$ </tex-math></inline-formula>\u0000) and (\u0000<inline-formula> <tex-math>$127~pm ~3$ </tex-math></inline-formula>\u0000) ps FWHM. We employed FastIC with a TlCl pure Cherenkov emitter, demonstrating time resolutions comparable to those achieved with the high-power-consuming electronics. These findings shows that the FastIC represents a cost-effective alternative that can significantly enhance the time resolution of the current TOF-PET systems while maintaining low power consumption.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"718-733"},"PeriodicalIF":4.6,"publicationDate":"2024-06-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10557761","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143649","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The PET/MRI system plays a significant role in the functional and anatomical quantification of the brain, providing accurate diagnostic data for a variety of brain disorders. However, most of the current methods for segmenting the brain are based on unimodal MRI and rarely combine structural and functional dual-modality information. Therefore, we aimed to employ deep-learning techniques to achieve automatic and accurate segmentation of the whole brain while incorporating functional and anatomical information. To leverage dual-modality information, a novel 3-D network with a cross-attention module was proposed to capture the correlation between dual-modality features and improve segmentation accuracy. Moreover, several deep-learning methods were employed as comparison measures to evaluate the model performance, with the dice similarity coefficient (DSC), Jaccard index (JAC), recall, and precision serving as quantitative metrics. Experimental results demonstrated our advantages in whole-brain segmentation, achieving an 85.35% DSC, 77.22% JAC, 88.86% recall, and 84.81% precision, which were better than those comparative methods. In addition, consistent and correlated analyses based on segmentation results also demonstrated that our approach achieved superior performance. In future work, we will try to apply our method to other multimodal tasks, such as PET/CT data analysis.
{"title":"Accurate Whole-Brain Segmentation for Bimodal PET/MR Images via a Cross-Attention Mechanism","authors":"Wenbo Li;Zhenxing Huang;Qiyang Zhang;Na Zhang;Wenjie Zhao;Yaping Wu;Jianmin Yuan;Yang Yang;Yan Zhang;Yongfeng Yang;Hairong Zheng;Dong Liang;Meiyun Wang;Zhanli Hu","doi":"10.1109/TRPMS.2024.3413862","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3413862","url":null,"abstract":"The PET/MRI system plays a significant role in the functional and anatomical quantification of the brain, providing accurate diagnostic data for a variety of brain disorders. However, most of the current methods for segmenting the brain are based on unimodal MRI and rarely combine structural and functional dual-modality information. Therefore, we aimed to employ deep-learning techniques to achieve automatic and accurate segmentation of the whole brain while incorporating functional and anatomical information. To leverage dual-modality information, a novel 3-D network with a cross-attention module was proposed to capture the correlation between dual-modality features and improve segmentation accuracy. Moreover, several deep-learning methods were employed as comparison measures to evaluate the model performance, with the dice similarity coefficient (DSC), Jaccard index (JAC), recall, and precision serving as quantitative metrics. Experimental results demonstrated our advantages in whole-brain segmentation, achieving an 85.35% DSC, 77.22% JAC, 88.86% recall, and 84.81% precision, which were better than those comparative methods. In addition, consistent and correlated analyses based on segmentation results also demonstrated that our approach achieved superior performance. In future work, we will try to apply our method to other multimodal tasks, such as PET/CT data analysis.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 1","pages":"47-56"},"PeriodicalIF":4.6,"publicationDate":"2024-06-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142912440","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-11DOI: 10.1109/TRPMS.2024.3410092
Dennis Hein;Staffan Holmin;Timothy Szczykutowicz;Jonathan S. Maltz;Mats Danielsson;Ge Wang;Mats Persson
Diffusion and Poisson flow models have shown impressive performance in a wide range of generative tasks, including low-dose CT (LDCT) image denoising. However, one limitation in general, and for clinical applications in particular, is slow sampling. Due to their iterative nature, the number of function evaluations (NFEs) required is usually on the order of $10-10^{3}$ , both for conditional and unconditional generation. In this article, we present posterior sampling Poisson flow generative models (PPFMs), a novel image denoising technique for low-dose and photon-counting CT that produces excellent image quality whilst keeping NFE = 1. Updating the training and sampling processes of Poisson flow generative models (PFGMs)++, we learn a conditional generator which defines a trajectory between the prior noise distribution and the posterior distribution of interest. We additionally hijack and regularize the sampling process to achieve NFE = 1. Our results shed light on the benefits of the PFGM++ framework compared to diffusion models. In addition, PPFM is shown to perform favorably compared to current state-of-the-art diffusion-style models with NFE = 1, consistency models, as well as popular deep learning and nondeep learning-based image denoising techniques, on clinical LDCT images and clinical images from a prototype photon-counting CT system.
{"title":"PPFM: Image Denoising in Photon-Counting CT Using Single-Step Posterior Sampling Poisson Flow Generative Models","authors":"Dennis Hein;Staffan Holmin;Timothy Szczykutowicz;Jonathan S. Maltz;Mats Danielsson;Ge Wang;Mats Persson","doi":"10.1109/TRPMS.2024.3410092","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3410092","url":null,"abstract":"Diffusion and Poisson flow models have shown impressive performance in a wide range of generative tasks, including low-dose CT (LDCT) image denoising. However, one limitation in general, and for clinical applications in particular, is slow sampling. Due to their iterative nature, the number of function evaluations (NFEs) required is usually on the order of \u0000<inline-formula> <tex-math>$10-10^{3}$ </tex-math></inline-formula>\u0000, both for conditional and unconditional generation. In this article, we present posterior sampling Poisson flow generative models (PPFMs), a novel image denoising technique for low-dose and photon-counting CT that produces excellent image quality whilst keeping NFE = 1. Updating the training and sampling processes of Poisson flow generative models (PFGMs)++, we learn a conditional generator which defines a trajectory between the prior noise distribution and the posterior distribution of interest. We additionally hijack and regularize the sampling process to achieve NFE = 1. Our results shed light on the benefits of the PFGM++ framework compared to diffusion models. In addition, PPFM is shown to perform favorably compared to current state-of-the-art diffusion-style models with NFE = 1, consistency models, as well as popular deep learning and nondeep learning-based image denoising techniques, on clinical LDCT images and clinical images from a prototype photon-counting CT system.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"788-799"},"PeriodicalIF":4.6,"publicationDate":"2024-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10554640","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143752","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-11DOI: 10.1109/TRPMS.2024.3412798
Andrea Gonzalez-Montoro;Santiago Jiménez-Serrano;Jorge Álamo;Julio Barberá;Alejandro Lucero;Neus Cucarella;Karel Díaz;Marta Freire;Antonio J. Gonzalez;Laura Moliner;Álvaro Mondejar;Constantino Morera-Ballester;John Prior;David Sánchez;Jose M. Benlloch
Positron emission tomography (PET) imaging is the molecular technique of choice for studying many illnesses, including the ones related to the brain. Nevertheless, the use of PET scanners in neurology is limited by several factors, such as their limited availability for brain imaging due to the high oncology demand for PET and the low sensitivity and poor spatial resolution in the brain of the standard PET scanners. To expand the PET application in neurology, the brain-specific systems with increased clinical and physical sensitivities and higher spatial resolution are required. The present work reports on the design and development process of a compact dedicated PET scanner suitable for human brain imaging. This article includes the description and experimental validation of the detector components and their implementation in a full-size system called 4D-PET. The detector has been designed to simultaneously provide photon depth of interaction (DOI) and time of flight (TOF) information. It is based on the semi-monolithic LYSO modules optically coupled to silicon photomultipliers (SiPMs) and connected to a multiplexing readout. The analog output signals are fed to the PETsys TOFPET2 analog-specific integrated circuit circuits enabling scalability of the readout. The evaluation of the 4D-PET modules resulted in average detector resolutions of $2.1pm 1$ .0 mm, $3.4pm 1$ .8 mm, and $386pm 9$ ps for the y- (transaxial direction), DOI-, and coincidence time resolution TOF, respectively. The preliminary 4D-PET imaging performance is reported through the simulations and for the first time through the real reconstructed images (collected in the La Fe Hospital, Valencia).
正电子发射断层扫描（PET）成像是研究许多疾病（包括与脑有关的疾病）的首选分子技术。然而，PET 扫描仪在神经病学中的应用受到几个因素的限制，例如，由于肿瘤学对 PET 的需求很高，因此脑成像的可用性有限，以及标准 PET 扫描仪的灵敏度低、脑部空间分辨率差。为了扩大 PET 在神经学领域的应用，需要临床和物理灵敏度更高和空间分辨率更高的脑部专用系统。本研究报告介绍了适用于人脑成像的紧凑型专用 PET 扫描仪的设计和开发过程。这篇文章包括探测器组件的描述和实验验证，以及它们在名为 4D-PET 的全尺寸系统中的实施情况。探测器的设计目的是同时提供光子相互作用深度（DOI）和飞行时间（TOF）信息。它基于半单片式 LYSO 模块，与硅光电倍增管（SiPM）光学耦合，并连接到多路复用读出器。模拟输出信号被馈送到 PETsys TOFPET2 模拟专用集成电路电路，从而实现了读出的可扩展性。通过对 4D-PET 模块的评估，Y-（横轴方向）、DOI- 和重合时间分辨率 TOF 的平均探测器分辨率分别为 2.1/pm 1$ .0 mm、3.4/pm 1$ .8 mm 和 386/pm 9$ ps。通过模拟和首次通过真实重建图像（在巴伦西亚拉费医院采集）报告了初步的 4D-PET 成像性能。
{"title":"First Results of the 4D-PET Brain System","authors":"Andrea Gonzalez-Montoro;Santiago Jiménez-Serrano;Jorge Álamo;Julio Barberá;Alejandro Lucero;Neus Cucarella;Karel Díaz;Marta Freire;Antonio J. Gonzalez;Laura Moliner;Álvaro Mondejar;Constantino Morera-Ballester;John Prior;David Sánchez;Jose M. Benlloch","doi":"10.1109/TRPMS.2024.3412798","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3412798","url":null,"abstract":"Positron emission tomography (PET) imaging is the molecular technique of choice for studying many illnesses, including the ones related to the brain. Nevertheless, the use of PET scanners in neurology is limited by several factors, such as their limited availability for brain imaging due to the high oncology demand for PET and the low sensitivity and poor spatial resolution in the brain of the standard PET scanners. To expand the PET application in neurology, the brain-specific systems with increased clinical and physical sensitivities and higher spatial resolution are required. The present work reports on the design and development process of a compact dedicated PET scanner suitable for human brain imaging. This article includes the description and experimental validation of the detector components and their implementation in a full-size system called 4D-PET. The detector has been designed to simultaneously provide photon depth of interaction (DOI) and time of flight (TOF) information. It is based on the semi-monolithic LYSO modules optically coupled to silicon photomultipliers (SiPMs) and connected to a multiplexing readout. The analog output signals are fed to the PETsys TOFPET2 analog-specific integrated circuit circuits enabling scalability of the readout. The evaluation of the 4D-PET modules resulted in average detector resolutions of \u0000<inline-formula> <tex-math>$2.1pm 1$ </tex-math></inline-formula>\u0000.0 mm, \u0000<inline-formula> <tex-math>$3.4pm 1$ </tex-math></inline-formula>\u0000.8 mm, and \u0000<inline-formula> <tex-math>$386pm 9$ </tex-math></inline-formula>\u0000 ps for the y- (transaxial direction), DOI-, and coincidence time resolution TOF, respectively. The preliminary 4D-PET imaging performance is reported through the simulations and for the first time through the real reconstructed images (collected in the La Fe Hospital, Valencia).","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"839-849"},"PeriodicalIF":4.6,"publicationDate":"2024-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10554551","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143720","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-06DOI: 10.1109/TRPMS.2024.3408870
Mikiko Ito;Dahea Han;Tae-Hyung Kim;Young-Tae Kim;Sungeun Lee;Jeongtae Soh;Young-Jun Jung;Byungkee Lee
Digital tomosynthesis (DTS) can enhance diagnostic accuracy by providing 3-D volume images with a remarkably low-X-ray dose. The aim of this study is to provide an initial assessment of the image quality and the X-ray dose for a mobile DTS system employing a moving carbon-nanotube (CNT)-based digital X-ray source array and a fixed detector for extremity scans. This design allows to reduce the source-to-detector distance (SDD) to only 400 mm, thereby enabling a compact and highly mobile system. We first measured the entrance surface dose (ESD), which is the sum of the X-ray dose irradiated from individual projections using a dosimeter placed at the center of the X-ray detector. The ESDs obtained for hand, foot, and knee scan configurations were 0.15, 0.22, and 0.43 mGy, respectively, which were comparable to those obtained from 2-D radiography exposures. For the evaluation of its reconstructed image quality, the in-plane modulation transfer function (MTF), Z -resolution, geometry distortion, and image homogeneity were assessed by utilizing a wire-phantom, sphere-phantom, and PMMA phantoms. The reconstructed images of hand, ankle and knee phantoms were evaluated qualitatively. The results of the evaluation demonstrate the successful development of the mobile DTS system proposed in this article.
数字断层扫描（DTS）能以极低的 X 射线剂量提供三维容积图像，从而提高诊断的准确性。本研究的目的是对采用移动碳纳米管（CNT）数字 X 射线源阵列和固定探测器进行四肢扫描的移动 DTS 系统的图像质量和 X 射线剂量进行初步评估。这种设计可将光源到探测器的距离（SDD）缩短到仅 400 毫米，从而实现了系统的紧凑性和高度移动性。我们首先测量了入口表面剂量（ESD），即使用放置在 X 射线探测器中心的剂量计测量从单个投影照射的 X 射线剂量的总和。手部、足部和膝部扫描配置获得的 ESD 分别为 0.15、0.22 和 0.43 mGy，与二维放射摄影曝光获得的 ESD 相当。为了评估其重建图像的质量，利用线状模型、球状模型和 PMMA 模型对平面内调制传递函数（MTF）、Z 分辨率、几何失真和图像均匀性进行了评估。对手部、踝关节和膝关节模型的重建图像进行了定性评估。评估结果表明，本文提出的移动 DTS 系统开发成功。
{"title":"Performance Evaluation of a Mobile Digital Tomosynthesis System Using a Moving CNT-Based Tube Array for Extremity Scans","authors":"Mikiko Ito;Dahea Han;Tae-Hyung Kim;Young-Tae Kim;Sungeun Lee;Jeongtae Soh;Young-Jun Jung;Byungkee Lee","doi":"10.1109/TRPMS.2024.3408870","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3408870","url":null,"abstract":"Digital tomosynthesis (DTS) can enhance diagnostic accuracy by providing 3-D volume images with a remarkably low-X-ray dose. The aim of this study is to provide an initial assessment of the image quality and the X-ray dose for a mobile DTS system employing a moving carbon-nanotube (CNT)-based digital X-ray source array and a fixed detector for extremity scans. This design allows to reduce the source-to-detector distance (SDD) to only 400 mm, thereby enabling a compact and highly mobile system. We first measured the entrance surface dose (ESD), which is the sum of the X-ray dose irradiated from individual projections using a dosimeter placed at the center of the X-ray detector. The ESDs obtained for hand, foot, and knee scan configurations were 0.15, 0.22, and 0.43 mGy, respectively, which were comparable to those obtained from 2-D radiography exposures. For the evaluation of its reconstructed image quality, the in-plane modulation transfer function (MTF), \u0000<italic>Z</i>\u0000-resolution, geometry distortion, and image homogeneity were assessed by utilizing a wire-phantom, sphere-phantom, and PMMA phantoms. The reconstructed images of hand, ankle and knee phantoms were evaluated qualitatively. The results of the evaluation demonstrate the successful development of the mobile DTS system proposed in this article.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"826-838"},"PeriodicalIF":4.6,"publicationDate":"2024-06-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143750","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-05DOI: 10.1109/TRPMS.2024.3410021
Jiayi Lu;Renchao Jin;Enmin Song
Objective: Serial medical image registration plays an important role in radiation therapy treatment planning. However, current deep learning-based deformable registration models suffer from excessive resource consumption and suboptimal precision issues. Moreover, the global regularization term may result in unrealistic deformations due to displacement field noise and intertissue sliding motion omission. Methods: This article proposes a patch-based pyramid convolutional recurrent neural network (pyramid CRNet) for serial medical image registration. Patch-wise training is employed to alleviate resource constraints. Incorporating spatiotemporal features across multiple scales is beneficial for focusing on more details to improve accuracy. Moreover, two motion adaptive techniques are introduced to provide anatomically plausible displacement fields. The first uses a guided filter to reduce noise and maintain motion continuity within organs. The second involves a pixel-wise weight regularization term within the loss function to provide a tailored solution for distinctive tissue characteristics, especially for sliding motion at organ boundaries. Results: Experiments were conducted on lung 4DCT images and cardiac cine MR images. Quantitative and qualitative results have demonstrated that our method can align anatomical structures across multiple images in a physiologically sensible manner. Conclusion: The significance of this work lies in its potential to address pressing challenges in clinical applications, and further investigations could be extended to explore different modalities and dimensions.
{"title":"Pyramid Convolutional Recurrent Network for Serial Medical Image Registration With Adaptive Motion Regularizations","authors":"Jiayi Lu;Renchao Jin;Enmin Song","doi":"10.1109/TRPMS.2024.3410021","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3410021","url":null,"abstract":"<italic>Objective:</i>\u0000 Serial medical image registration plays an important role in radiation therapy treatment planning. However, current deep learning-based deformable registration models suffer from excessive resource consumption and suboptimal precision issues. Moreover, the global regularization term may result in unrealistic deformations due to displacement field noise and intertissue sliding motion omission. \u0000<italic>Methods:</i>\u0000 This article proposes a patch-based pyramid convolutional recurrent neural network (pyramid CRNet) for serial medical image registration. Patch-wise training is employed to alleviate resource constraints. Incorporating spatiotemporal features across multiple scales is beneficial for focusing on more details to improve accuracy. Moreover, two motion adaptive techniques are introduced to provide anatomically plausible displacement fields. The first uses a guided filter to reduce noise and maintain motion continuity within organs. The second involves a pixel-wise weight regularization term within the loss function to provide a tailored solution for distinctive tissue characteristics, especially for sliding motion at organ boundaries. \u0000<italic>Results:</i>\u0000 Experiments were conducted on lung 4DCT images and cardiac cine MR images. Quantitative and qualitative results have demonstrated that our method can align anatomical structures across multiple images in a physiologically sensible manner. \u0000<italic>Conclusion:</i>\u0000 The significance of this work lies in its potential to address pressing challenges in clinical applications, and further investigations could be extended to explore different modalities and dimensions.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"800-813"},"PeriodicalIF":4.6,"publicationDate":"2024-06-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143653","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-06-03DOI: 10.1109/TRPMS.2024.3408818
Yuya Kuriyama;Mitsuhiro Nakamura;Megumi Nakao
The field of medical machine learning has encountered the challenge of constructing a large-scale image database that includes both the anatomical variability and teaching labels because there are often not sufficient cases of a specific disease. Adversarial learning has been studied for nonlinear data augmentation. However, deep learning models may produce anatomically unrealistic structures or inaccurate pixel values when applied to small sets of computed tomography (CT) images. To overcome this issue, we propose a data augmentation method that uses the hierarchical encoding of deformation fields between the CT images. This allows for the generation of synthetic CT images with shape variability while preserving the patient-specific CT values. Our framework encodes the spatial features of deformation fields into hierarchical latent variables, and generates the synthetic deformation fields by updating the values in specific layers. To implement this concept, we applied the StyleGAN2 and its encoder pixel2style2pixel to the deformation fields and added the ability to control the level of detail in the deformation through the Style Mixing. Our experiments demonstrated that our framework produced high-quality synthetic CT images compared with a conventional framework. Additionally, we applied the augmented datasets with teaching labels to semantic segmentation tasks targeting the liver and stomach, and found that accuracy improved by 1.3% and 7.9%, respectively, which surpassed the results obtained by the existing data augmentation methods.
{"title":"Data Augmentation Using the Hierarchical Encoding of Deformation Fields Between CT Images","authors":"Yuya Kuriyama;Mitsuhiro Nakamura;Megumi Nakao","doi":"10.1109/TRPMS.2024.3408818","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3408818","url":null,"abstract":"The field of medical machine learning has encountered the challenge of constructing a large-scale image database that includes both the anatomical variability and teaching labels because there are often not sufficient cases of a specific disease. Adversarial learning has been studied for nonlinear data augmentation. However, deep learning models may produce anatomically unrealistic structures or inaccurate pixel values when applied to small sets of computed tomography (CT) images. To overcome this issue, we propose a data augmentation method that uses the hierarchical encoding of deformation fields between the CT images. This allows for the generation of synthetic CT images with shape variability while preserving the patient-specific CT values. Our framework encodes the spatial features of deformation fields into hierarchical latent variables, and generates the synthetic deformation fields by updating the values in specific layers. To implement this concept, we applied the StyleGAN2 and its encoder pixel2style2pixel to the deformation fields and added the ability to control the level of detail in the deformation through the Style Mixing. Our experiments demonstrated that our framework produced high-quality synthetic CT images compared with a conventional framework. Additionally, we applied the augmented datasets with teaching labels to semantic segmentation tasks targeting the liver and stomach, and found that accuracy improved by 1.3% and 7.9%, respectively, which surpassed the results obtained by the existing data augmentation methods.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 8","pages":"939-949"},"PeriodicalIF":4.6,"publicationDate":"2024-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142587626","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
High-performance radiation detectors often employ crystal arrays where light can leak between them, a phenomenon called intercrystal crosstalk, which demands mitigation for optimal detector performance. The complexity of measuring optical crosstalk in conventional detector geometries makes optical Monte Carlo simulation essential to study and reduce crosstalk through better designs. Addressing the absence of validated transmission models in Monte Carlo toolkits, we developed and integrated a new simulation model into the look-up table Davis Model, aiming at simulating optical photon refraction at the crystal interfaces using GATE. For the first time, we validated the intercrystal optical crosstalk model with experiments in two optically coupled Lutetium-yttrium oxyorthosilicate crystals read by two SiPMs, testing three thicknesses and four interfaces (air, glue, Teflon, and ESR). Simulated and experimental crosstalk agreed within one FWHM for all configurations. These results show the possibility of predicting optical photon transmission in detector designs with multiple crystal elements. Indeed, although validated using only two crystals, the model can be used in more complex geometries. The model, available to GATE users upon request, provides a valuable resource for researchers when optimizing detector geometry where optical crosstalk needs to be considered, i.e., ensuring optical isolation between the photodetector’s responses.
{"title":"Intercrystal Optical Crosstalk in Radiation Detectors: Monte Carlo Modeling and Experimental Validation","authors":"Carlotta Trigila;N. Kratochwil;B. Mehadji;G. Ariño-Estrada;E. Roncali","doi":"10.1109/TRPMS.2024.3395131","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3395131","url":null,"abstract":"High-performance radiation detectors often employ crystal arrays where light can leak between them, a phenomenon called intercrystal crosstalk, which demands mitigation for optimal detector performance. The complexity of measuring optical crosstalk in conventional detector geometries makes optical Monte Carlo simulation essential to study and reduce crosstalk through better designs. Addressing the absence of validated transmission models in Monte Carlo toolkits, we developed and integrated a new simulation model into the look-up table Davis Model, aiming at simulating optical photon refraction at the crystal interfaces using GATE. For the first time, we validated the intercrystal optical crosstalk model with experiments in two optically coupled Lutetium-yttrium oxyorthosilicate crystals read by two SiPMs, testing three thicknesses and four interfaces (air, glue, Teflon, and ESR). Simulated and experimental crosstalk agreed within one FWHM for all configurations. These results show the possibility of predicting optical photon transmission in detector designs with multiple crystal elements. Indeed, although validated using only two crystals, the model can be used in more complex geometries. The model, available to GATE users upon request, provides a valuable resource for researchers when optimizing detector geometry where optical crosstalk needs to be considered, i.e., ensuring optical isolation between the photodetector’s responses.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"734-742"},"PeriodicalIF":4.6,"publicationDate":"2024-04-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10510415","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143679","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-04-24DOI: 10.1109/TRPMS.2024.3393235
Ran Cheng;Mingchen Sun;Fei Wang;Dengyun Mu;Yu Liu;Qingguo Xie;Bensheng Qiu;Xun Chen;Peng Xiao
To minimize parallax errors and achieve high spatial resolution positron emission tomography (PET) systems, developing depth-of-Interaction (DOI) encoding detectors has become a significant research topic. In this article, we investigated a dual-ended readout PET detector based on the multivoltage threshold (MVT) sampling method combined with a convolutional neural network (CNN) to calculate the pulse’s energy (MVT-CNN method). The MVT sampling method was used to acquire time-threshold samples and digitize scintillation pulses. The CNN model was employed to establish an accurate mapping between MVT sampling points and energy information. The dual-ended readout detector’s energy, DOI, and timing performance were evaluated with two irradiation configurations. The results demonstrated that the performance of the MVT-CNN method was close to that of the integration method based on oscilloscope sampling. Using the MVT-CNN method, the average energy resolution of the tested crystals over all depths was $14.5 , pm , 1.2$ %, and the average DOI resolution was $2.81 , pm , 0$ .70 mm. In the side irradiation configuration, the average coincidence timing resolution of the tested crystals at 2 mm depth was 435 ps. The performance of the dual-ended readout DOI-PET detector basedon the MVT-CNN method suggested that it could develop small animal and organ-dedicated PET systems with high sensitivity and uniform spatial resolutionxs.
{"title":"Dual-Ended Readout PET Detector Based on Multivoltage Threshold Sampling Combined With Convolutional Neural Network for Energy Calculation","authors":"Ran Cheng;Mingchen Sun;Fei Wang;Dengyun Mu;Yu Liu;Qingguo Xie;Bensheng Qiu;Xun Chen;Peng Xiao","doi":"10.1109/TRPMS.2024.3393235","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3393235","url":null,"abstract":"To minimize parallax errors and achieve high spatial resolution positron emission tomography (PET) systems, developing depth-of-Interaction (DOI) encoding detectors has become a significant research topic. In this article, we investigated a dual-ended readout PET detector based on the multivoltage threshold (MVT) sampling method combined with a convolutional neural network (CNN) to calculate the pulse’s energy (MVT-CNN method). The MVT sampling method was used to acquire time-threshold samples and digitize scintillation pulses. The CNN model was employed to establish an accurate mapping between MVT sampling points and energy information. The dual-ended readout detector’s energy, DOI, and timing performance were evaluated with two irradiation configurations. The results demonstrated that the performance of the MVT-CNN method was close to that of the integration method based on oscilloscope sampling. Using the MVT-CNN method, the average energy resolution of the tested crystals over all depths was \u0000<inline-formula> <tex-math>$14.5 , pm , 1.2$ </tex-math></inline-formula>\u0000%, and the average DOI resolution was \u0000<inline-formula> <tex-math>$2.81 , pm , 0$ </tex-math></inline-formula>\u0000.70 mm. In the side irradiation configuration, the average coincidence timing resolution of the tested crystals at 2 mm depth was 435 ps. The performance of the dual-ended readout DOI-PET detector basedon the MVT-CNN method suggested that it could develop small animal and organ-dedicated PET systems with high sensitivity and uniform spatial resolutionxs.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"709-717"},"PeriodicalIF":4.6,"publicationDate":"2024-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143651","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：