X-ray imaging is crucial in orthopedic disease detection and diagnosis, but it can impact the body significantly. Ensuring imaging quality is vital for accurate diagnoses and reducing repeat scans. However, quality inspection can decrease efficiency and be influenced by subjectivity when handling large data volumes, affecting evaluation outcomes. Current deep learning methods for medical image quality assessment rely on extensive labeled data, posing privacy and resource challenges. Our research aims to develop a quality assessment network for X-ray imaging independent of complex labels and large datasets, tailored for multi-index quality assessment. We propose an X-ray imaging quality assessment network based on segmentation priors, utilizing the “segment anything model” (SAM) for mask segmentation and a dual-feature extraction network to process prior information. Through a channel fully connected module, we transform the regression problem into a multiclassification problem, improving convergence speed and performance. Comparative analysis demonstrates the superiority of our proposed algorithm. Our X-ray imaging quality assessment network achieves accurate and efficient quality assessment without relying on extensive labeled data. https://github.com/OPMZZZ/SAM-DRIQA/
{"title":"Segmentation-Based X-Ray Multiobjective Quality Assessment Network","authors":"Qianyi Yang;Demin Xu;Zhenxing Huang;Wenbo Li;Guanxun Cheng;Tianye Niu;Hairong Zheng;Dong Liang;Fei Feng;Zhanli Hu","doi":"10.1109/TRPMS.2024.3452683","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3452683","url":null,"abstract":"X-ray imaging is crucial in orthopedic disease detection and diagnosis, but it can impact the body significantly. Ensuring imaging quality is vital for accurate diagnoses and reducing repeat scans. However, quality inspection can decrease efficiency and be influenced by subjectivity when handling large data volumes, affecting evaluation outcomes. Current deep learning methods for medical image quality assessment rely on extensive labeled data, posing privacy and resource challenges. Our research aims to develop a quality assessment network for X-ray imaging independent of complex labels and large datasets, tailored for multi-index quality assessment. We propose an X-ray imaging quality assessment network based on segmentation priors, utilizing the “segment anything model” (SAM) for mask segmentation and a dual-feature extraction network to process prior information. Through a channel fully connected module, we transform the regression problem into a multiclassification problem, improving convergence speed and performance. Comparative analysis demonstrates the superiority of our proposed algorithm. Our X-ray imaging quality assessment network achieves accurate and efficient quality assessment without relying on extensive labeled data. <uri>https://github.com/OPMZZZ/SAM-DRIQA/</uri>","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 2","pages":"202-214"},"PeriodicalIF":4.6,"publicationDate":"2024-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143106267","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 : 2024-09-02DOI: 10.1109/TRPMS.2024.3453401
Yuxiang Yang;Xinyi Zeng;Pinxian Zeng;Binyu Yan;Xi Wu;Jiliu Zhou;Yan Wang
Deep learning has revolutionized the early detection of breast cancer, resulting in a significant decrease in mortality rates. However, difficulties in obtaining annotations and huge variations in distribution between training sets and real scenes have limited their clinical applications. To address these limitations, unsupervised domain adaptation (UDA) methods have been used to transfer knowledge from one labeled source domain to the unlabeled target domain, yet these approaches suffer from severe domain shift issues and often ignore the potential benefits of leveraging multiple relevant sources in practical applications. To address these limitations, in this work, we construct a three-branch mixed extractor and propose a bi-level multisource UDA method called BTMuda for breast cancer diagnosis. Our method addresses the problems of domain shift by dividing domain shift issues into two levels: 1) intradomain and 2) interdomain. To reduce the intradomain shift, we jointly train a convolutional neural network and a Transformer as two paths of a domain mixed feature extractor to obtain robust representations rich in both low-level local and high-level global information. As for the interdomain shift, we redesign the Transformer delicately to a three-branch architecture with cross-attention and distillation, which learns domain-invariant representations from multiple domains. Besides, we introduce two alignment modules—one for feature alignment and one for classifier alignment—to improve the alignment process. Extensive experiments conducted on three public mammographic datasets demonstrate that our BTMuda outperforms state-of-the-art methods.
{"title":"BTMuda: A Bi-Level Multisource Unsupervised Domain Adaptation Framework for Breast Cancer Diagnosis","authors":"Yuxiang Yang;Xinyi Zeng;Pinxian Zeng;Binyu Yan;Xi Wu;Jiliu Zhou;Yan Wang","doi":"10.1109/TRPMS.2024.3453401","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3453401","url":null,"abstract":"Deep learning has revolutionized the early detection of breast cancer, resulting in a significant decrease in mortality rates. However, difficulties in obtaining annotations and huge variations in distribution between training sets and real scenes have limited their clinical applications. To address these limitations, unsupervised domain adaptation (UDA) methods have been used to transfer knowledge from one labeled source domain to the unlabeled target domain, yet these approaches suffer from severe domain shift issues and often ignore the potential benefits of leveraging multiple relevant sources in practical applications. To address these limitations, in this work, we construct a three-branch mixed extractor and propose a bi-level multisource UDA method called BTMuda for breast cancer diagnosis. Our method addresses the problems of domain shift by dividing domain shift issues into two levels: 1) intradomain and 2) interdomain. To reduce the intradomain shift, we jointly train a convolutional neural network and a Transformer as two paths of a domain mixed feature extractor to obtain robust representations rich in both low-level local and high-level global information. As for the interdomain shift, we redesign the Transformer delicately to a three-branch architecture with cross-attention and distillation, which learns domain-invariant representations from multiple domains. Besides, we introduce two alignment modules—one for feature alignment and one for classifier alignment—to improve the alignment process. Extensive experiments conducted on three public mammographic datasets demonstrate that our BTMuda outperforms state-of-the-art methods.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 3","pages":"313-324"},"PeriodicalIF":4.6,"publicationDate":"2024-09-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10663460","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553042","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 : 2024-08-28DOI: 10.1109/TRPMS.2024.3450833
Emily Anaya;Paul Schleyer;Craig Levin
In simultaneous positron emission tomography and magnetic resonance (PET/MR) imaging, MR radio-frequency (RF) coils are placed on the top of the patient to receive the MR signal. These coils can produce an undesirable photon attenuation of the PET signal by as much as 17% in certain local regions of a reconstructed PET cylindrical phantom. Currently, photon attenuation of RF body coils is not typically accounted for in the attenuation correction (AC) procedure in commercial PET/MR systems. To correct for this coil attenuation, the position of the coils and their most attenuating components, such as the preamplifier housings must be accurately determined. This work proposes a simple and effective solution to this problem by using three optical cameras placed just outside the field-of-view (FOV) of the PET/MR system. The cameras are used to determine the positions of markers attached to the RF coils. An average marker location error of 7.7 mm was achieved over eight markers placed on a flexible RF coil draped over a cylindrical PET phantom. Quantification of reconstructed PET signal error due to inaccurate assessment of flexible RF coil location on a phantom is presented. Given the coil location accuracy of this method, the PET signal attenuation error is reduced from 17% to less than 3%. Our method can also be extended to correct for other attenuating objects in the FOV of the PET/MR system.
{"title":"A Method to Locate Radio-Frequency Coils Using a CT-Based Template for a More Accurate Photon Attenuation Correction in PET/MRI","authors":"Emily Anaya;Paul Schleyer;Craig Levin","doi":"10.1109/TRPMS.2024.3450833","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3450833","url":null,"abstract":"In simultaneous positron emission tomography and magnetic resonance (PET/MR) imaging, MR radio-frequency (RF) coils are placed on the top of the patient to receive the MR signal. These coils can produce an undesirable photon attenuation of the PET signal by as much as 17% in certain local regions of a reconstructed PET cylindrical phantom. Currently, photon attenuation of RF body coils is not typically accounted for in the attenuation correction (AC) procedure in commercial PET/MR systems. To correct for this coil attenuation, the position of the coils and their most attenuating components, such as the preamplifier housings must be accurately determined. This work proposes a simple and effective solution to this problem by using three optical cameras placed just outside the field-of-view (FOV) of the PET/MR system. The cameras are used to determine the positions of markers attached to the RF coils. An average marker location error of 7.7 mm was achieved over eight markers placed on a flexible RF coil draped over a cylindrical PET phantom. Quantification of reconstructed PET signal error due to inaccurate assessment of flexible RF coil location on a phantom is presented. Given the coil location accuracy of this method, the PET signal attenuation error is reduced from 17% to less than 3%. Our method can also be extended to correct for other attenuating objects in the FOV of the PET/MR system.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 2","pages":"182-190"},"PeriodicalIF":4.6,"publicationDate":"2024-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143106269","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}
4-Dcone-beam computed tomography (4-D CBCT) has recently been recognized as a proficient technique in mitigating motion artifacts attributed to respiratory organ movement. The primary challenges in 4-D CBCT reconstruction encompass the precision in projection grouping, the efficacy in reconstructing from sparsely sampled data, and the accuracy in deformation field estimation. To surmount these challenges, we propose an innovative approach that integrates meticulous respiratory curve extraction for projection grouping and utilizes a diffusion model network with motion compensation (MoCo) techniques targeted at significantly enhancing image quality. An object detection network is employed to ascertain the exact position of the diaphragm, which is then normalized to formulate the respiratory curve. Further, we employ a U-Net architecture-based diffusion model, which integrates attention mechanisms to enhance sparse-view reconstruction and reduce artifacts through Guided-Diffusion. Deviating from conventional optical flow methods, our approach introduces an unsupervised registration network for deformation vector field (DVF) in phase-enhanced images. This DVF is then utilized in a motion-compensated, ordered-subset, simultaneous algebraic reconstruction technique, culminating in the generation of 4-D CBCT images. The efficacy of this method has been substantiated through validation on both simulated and clinical datasets, with the results from comparative experiments indicating promising outcomes.
{"title":"4-D Cone-Beam CT Reconstruction via Diffusion Model and Motion Compensation","authors":"Xianghong Wang;Zhengwei Ou;Peng Jin;Jiayi Xie;Ze Teng;Lei Xu;Jichen Du;Mingchao Ding;Yang Chen;Tianye Niu","doi":"10.1109/TRPMS.2024.3449155","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3449155","url":null,"abstract":"4-Dcone-beam computed tomography (4-D CBCT) has recently been recognized as a proficient technique in mitigating motion artifacts attributed to respiratory organ movement. The primary challenges in 4-D CBCT reconstruction encompass the precision in projection grouping, the efficacy in reconstructing from sparsely sampled data, and the accuracy in deformation field estimation. To surmount these challenges, we propose an innovative approach that integrates meticulous respiratory curve extraction for projection grouping and utilizes a diffusion model network with motion compensation (MoCo) techniques targeted at significantly enhancing image quality. An object detection network is employed to ascertain the exact position of the diaphragm, which is then normalized to formulate the respiratory curve. Further, we employ a U-Net architecture-based diffusion model, which integrates attention mechanisms to enhance sparse-view reconstruction and reduce artifacts through Guided-Diffusion. Deviating from conventional optical flow methods, our approach introduces an unsupervised registration network for deformation vector field (DVF) in phase-enhanced images. This DVF is then utilized in a motion-compensated, ordered-subset, simultaneous algebraic reconstruction technique, culminating in the generation of 4-D CBCT images. The efficacy of this method has been substantiated through validation on both simulated and clinical datasets, with the results from comparative experiments indicating promising outcomes.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 2","pages":"191-201"},"PeriodicalIF":4.6,"publicationDate":"2024-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10644124","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143106266","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}
Nonthermal plasma, cold plasma, and atmospheric-pressure plasma are few terms used to describe the plasma used in plasma medicine research. The resulting ambiguity hampers literature searches, confuses discussion, and complicates collaborations. To assess the full breadth of this problem, we designed a natural language processing (NLP) model that surveyed approximately 15 000 papers in response to the query “plasma medicine” indexed in PubMed between 2020 and 2022. Our NLP was constructed and executed using the Hugging Face transformers API and PubMed BERT pretrained model. We used this model to determine the prevalence and to assess the utility of each term for searching literature relevant to plasma medicine. The effectiveness of each term was measured by precision, the ability to discriminate relevant and irrelevant literature; and recall, the ability to retrieve relevant literature. Each term was given a combined effectiveness score of 0-1 ($1{=}$ ideal effectiveness) accounting for precision, recall, sample size, and model confidence. Our model showed that of the 12 commonly used terms analyzed, none received a combined effectiveness score over 0.025. We concluded that there is no universal term for “plasma” that provides a satisfactory representation of literature. These results highlight the need for standardization of nomenclature in plasma medicine.
{"title":"By Any Other Name: Searching for the Right Plasma Nomenclature","authors":"Caroline Corcoran;Rachel Bennett;Vandana Miller;Fred Krebs;Will Dampier","doi":"10.1109/TRPMS.2024.3447551","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3447551","url":null,"abstract":"Nonthermal plasma, cold plasma, and atmospheric-pressure plasma are few terms used to describe the plasma used in plasma medicine research. The resulting ambiguity hampers literature searches, confuses discussion, and complicates collaborations. To assess the full breadth of this problem, we designed a natural language processing (NLP) model that surveyed approximately 15 000 papers in response to the query “plasma medicine” indexed in PubMed between 2020 and 2022. Our NLP was constructed and executed using the Hugging Face transformers API and PubMed BERT pretrained model. We used this model to determine the prevalence and to assess the utility of each term for searching literature relevant to plasma medicine. The effectiveness of each term was measured by precision, the ability to discriminate relevant and irrelevant literature; and recall, the ability to retrieve relevant literature. Each term was given a combined effectiveness score of 0-1 (<inline-formula> <tex-math>$1{=}$ </tex-math></inline-formula> ideal effectiveness) accounting for precision, recall, sample size, and model confidence. Our model showed that of the 12 commonly used terms analyzed, none received a combined effectiveness score over 0.025. We concluded that there is no universal term for “plasma” that provides a satisfactory representation of literature. These results highlight the need for standardization of nomenclature in plasma medicine.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 3","pages":"388-394"},"PeriodicalIF":4.6,"publicationDate":"2024-08-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553284","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}
Hyperspectral unmixing aims at decomposing a given signal into its spectral signatures and its associated fractional abundances. To improve the accuracy of this decomposition, algorithms have included different assumptions depending on the application. The goal of this study is to develop a new unmixing algorithm that can be applied for the calibration of multipoint scintillation dosimeters used in the field of radiation therapy. This new algorithm is based on a non-negative matrix factorization. It incorporates a partial prior knowledge on both the abundances and the endmembers of a given signal. It is shown herein that, following a precise calibration routine, it is possible to use partial prior information about the fractional abundances, as well as on the endmembers, in order to perform a simplified yet precise calibration of these dosimeters. Validation and characterization of this algorithm is made using both simulations and experiments. The experimental validation shows an improvement in accuracy compared to previous algorithms with a mean spectral angle distance (SAD) on the estimated endmembers of 0.0766, leading to an average error of $(0.25 pm 0.73)$ % on dose measurements.
{"title":"Non-Negative Matrix Factorization Using Partial Prior Knowledge for Radiation Dosimetry","authors":"Boby Lessard;Frédéric Marcotte;Arthur Lalonde;François Therriault-Proulx;Simon Lambert-Girard;Luc Beaulieu;Louis Archambault","doi":"10.1109/TRPMS.2024.3442773","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3442773","url":null,"abstract":"Hyperspectral unmixing aims at decomposing a given signal into its spectral signatures and its associated fractional abundances. To improve the accuracy of this decomposition, algorithms have included different assumptions depending on the application. The goal of this study is to develop a new unmixing algorithm that can be applied for the calibration of multipoint scintillation dosimeters used in the field of radiation therapy. This new algorithm is based on a non-negative matrix factorization. It incorporates a partial prior knowledge on both the abundances and the endmembers of a given signal. It is shown herein that, following a precise calibration routine, it is possible to use partial prior information about the fractional abundances, as well as on the endmembers, in order to perform a simplified yet precise calibration of these dosimeters. Validation and characterization of this algorithm is made using both simulations and experiments. The experimental validation shows an improvement in accuracy compared to previous algorithms with a mean spectral angle distance (SAD) on the estimated endmembers of 0.0766, leading to an average error of <inline-formula> <tex-math>$(0.25 pm 0.73)$ </tex-math></inline-formula>% on dose measurements.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 2","pages":"247-258"},"PeriodicalIF":4.6,"publicationDate":"2024-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143106263","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 : 2024-08-16DOI: 10.1109/TRPMS.2024.3443831
Xiang Zhang;Yonggang Wang;Mingchen Wang;Xiaoguang Kong
Field programmable logic array (FPGA)-based readout electronics has shown its capability of channel-by-channel signal readout for time-of-flight positron emission tomography (TOF-PET) detectors. However, for detectors that rely on light sharing to achieve subpixel resolution, the high-linear measurement dynamic range of the readout electronics is highly required. In this article, the problems with dynamic range in our previously proposed FPGA-based fast linear discharge circuit are investigated and corresponding methods are proposed to enhance its small signal measurement capability and improve the timing performance as well. A practical 64-channel TOF-PET detector module was constructed and evaluated. The readout electronics test results demonstrated a 240x measurement dynamic range with 99.5% conversion linearity. In the case that the $8times 8$ silicon photomultiplier (SiPM) array in the detector combines with an $8times 8$ LYSO crystal (each $3.2times 3.2times 10$ mm3) array, the average energy and coincidence time resolution of the detector are measured as 10.68% (511 keV) and 364.9 ps, respectively. To demonstrate the benefit of large dynamic range to high-resolution detectors, the crystal array in the detector was replaced by a $24times 24$ LYSO array (each $1.04times 1.04times 15$ mm3) and achieved 1-mm resolution. The test results confirm that the proposed FPGA-based readout circuit is practical for laboratory instrumentation
{"title":"An FPGA-Based 64-Channel Readout Electronics for High-Resolution TOF-PET Detectors","authors":"Xiang Zhang;Yonggang Wang;Mingchen Wang;Xiaoguang Kong","doi":"10.1109/TRPMS.2024.3443831","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3443831","url":null,"abstract":"Field programmable logic array (FPGA)-based readout electronics has shown its capability of channel-by-channel signal readout for time-of-flight positron emission tomography (TOF-PET) detectors. However, for detectors that rely on light sharing to achieve subpixel resolution, the high-linear measurement dynamic range of the readout electronics is highly required. In this article, the problems with dynamic range in our previously proposed FPGA-based fast linear discharge circuit are investigated and corresponding methods are proposed to enhance its small signal measurement capability and improve the timing performance as well. A practical 64-channel TOF-PET detector module was constructed and evaluated. The readout electronics test results demonstrated a 240x measurement dynamic range with 99.5% conversion linearity. In the case that the \u0000<inline-formula> <tex-math>$8times 8$ </tex-math></inline-formula>\u0000 silicon photomultiplier (SiPM) array in the detector combines with an \u0000<inline-formula> <tex-math>$8times 8$ </tex-math></inline-formula>\u0000 LYSO crystal (each \u0000<inline-formula> <tex-math>$3.2times 3.2times 10$ </tex-math></inline-formula>\u0000 mm3) array, the average energy and coincidence time resolution of the detector are measured as 10.68% (511 keV) and 364.9 ps, respectively. To demonstrate the benefit of large dynamic range to high-resolution detectors, the crystal array in the detector was replaced by a \u0000<inline-formula> <tex-math>$24times 24$ </tex-math></inline-formula>\u0000 LYSO array (each \u0000<inline-formula> <tex-math>$1.04times 1.04times 15$ </tex-math></inline-formula>\u0000 mm3) and achieved 1-mm resolution. The test results confirm that the proposed FPGA-based readout circuit is practical for laboratory instrumentation","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 1","pages":"11-19"},"PeriodicalIF":4.6,"publicationDate":"2024-08-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142912439","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 : 2024-08-14DOI: 10.1109/TRPMS.2024.3439010
Ayush Chaturvedi;Ritvik Prabhu;Mukund Yadav;Wu-Chun Feng;Guohua Cao
Chest CT scans play an important role in diagnosing abnormalities associated with the lungs, such as tuberculosis, sarcoidosis, pneumonia, and, more recently, COVID-19. However, because conventional normal-dose chest CT scans require a much larger amount of radiation than x-rays, practitioners seek to replace conventional CT with low-dose CT (LDCT). LDCT often generates a low-quality CT image that poses noise and, in turn, negatively affects the accuracy of diagnosis. Therefore, in the context of COVID-19, due to the large number of affected populations, efficient image-denoising techniques are needed for LDCT images. Here, we present a deep learning (DL) model that combines two neural networks to enhance the quality of low-dose chest CT images. The DL model leverages a previously developed densenet and deconvolution-based network (DDNet) for feature extraction and extends it with a pretrained VGG network inside the loss function to suppress noise. Outputs from selected multiple levels in the VGG network (ML-VGG) are leveraged for the loss calculation. We tested our DDNet with ML-VGG loss using several sources of CT images and compared its performance to DDNet without VGG loss as well as DDNet with an empirically selected single-level VGG loss (DDNet-SL-VGG) and other state-of-the-art DL models. Our results show that DDNet combined with ML-VGG (DDNet-ML-VGG) achieves state-of-the-art denoising capabilities and improves the perceptual and quantitative image quality of chest CT images. Thus, DDNet with multilevel VGG loss could potentially be used as a post-acquisition image enhancement tool for medical professionals to diagnose and monitor chest diseases with higher accuracy.
{"title":"Improved 2-D Chest CT Image Enhancement With Multi-Level VGG Loss","authors":"Ayush Chaturvedi;Ritvik Prabhu;Mukund Yadav;Wu-Chun Feng;Guohua Cao","doi":"10.1109/TRPMS.2024.3439010","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3439010","url":null,"abstract":"Chest CT scans play an important role in diagnosing abnormalities associated with the lungs, such as tuberculosis, sarcoidosis, pneumonia, and, more recently, COVID-19. However, because conventional normal-dose chest CT scans require a much larger amount of radiation than x-rays, practitioners seek to replace conventional CT with low-dose CT (LDCT). LDCT often generates a low-quality CT image that poses noise and, in turn, negatively affects the accuracy of diagnosis. Therefore, in the context of COVID-19, due to the large number of affected populations, efficient image-denoising techniques are needed for LDCT images. Here, we present a deep learning (DL) model that combines two neural networks to enhance the quality of low-dose chest CT images. The DL model leverages a previously developed densenet and deconvolution-based network (DDNet) for feature extraction and extends it with a pretrained VGG network inside the loss function to suppress noise. Outputs from selected multiple levels in the VGG network (ML-VGG) are leveraged for the loss calculation. We tested our DDNet with ML-VGG loss using several sources of CT images and compared its performance to DDNet without VGG loss as well as DDNet with an empirically selected single-level VGG loss (DDNet-SL-VGG) and other state-of-the-art DL models. Our results show that DDNet combined with ML-VGG (DDNet-ML-VGG) achieves state-of-the-art denoising capabilities and improves the perceptual and quantitative image quality of chest CT images. Thus, DDNet with multilevel VGG loss could potentially be used as a post-acquisition image enhancement tool for medical professionals to diagnose and monitor chest diseases with higher accuracy.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 3","pages":"304-312"},"PeriodicalIF":4.6,"publicationDate":"2024-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553300","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 : 2024-08-12DOI: 10.1109/TRPMS.2024.3442690
Ju-Chieh Kevin Cheng;Erik Reimers;Vesna Sossi
We describe a deep learning (DL) assisted HYPR4D kernelized reconstruction which produces low-noise voxel-level time-activity-curves (TACs) while preserving quantification within small structures as well as consistent spatiotemporal patterns/features within measured data. The proposed method consists of the following advantages over other methods: 1) unsupervised single subject network training scheme independent of positron emission tomography (PET) tracers; 2) training data generated on-the-fly during reconstruction; 3) intrinsic spatiotemporal consistency provided by minimizing the $L_{2}$ loss using pseudo 4-D (i.e., 2.5 Spatial Dimension + 0.5 Temporal Dimension or 2.5SD+0.5TD) patches between kernelized OSEM subset estimates; and 4) a final tuning step which minimizes over-smoothing from the network output within the kernel matrix. Contrast phantom, human [18F]FDG and [11C]RAC data acquired on GE SIGNA PET/MR were used for evaluations. The proposed DL HYPR4D kernel method outperformed the standard HYPR4D kernel method as well as TOF-OSEM and TOF-BSREM (Q.Clear) in terms contrast recovery versus noise. The proposed final tuning reduced the underestimation bias due to over-smoothing within a 4-mm target structure from ~15% to ~2% while maintaining low-noise voxel-level TACs. In addition, the proposed unsupervised DL assisted reconstruction also outperformed the supervised DL version in terms of bias reduction along the TACs and kinetic model fittings.
{"title":"HYPR4D Kernel Method With an Unsupervised 2.5SD+0.5TD Deep Learning Assisted Kernel Matrix","authors":"Ju-Chieh Kevin Cheng;Erik Reimers;Vesna Sossi","doi":"10.1109/TRPMS.2024.3442690","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3442690","url":null,"abstract":"We describe a deep learning (DL) assisted HYPR4D kernelized reconstruction which produces low-noise voxel-level time-activity-curves (TACs) while preserving quantification within small structures as well as consistent spatiotemporal patterns/features within measured data. The proposed method consists of the following advantages over other methods: 1) unsupervised single subject network training scheme independent of positron emission tomography (PET) tracers; 2) training data generated on-the-fly during reconstruction; 3) intrinsic spatiotemporal consistency provided by minimizing the \u0000<inline-formula> <tex-math>$L_{2}$ </tex-math></inline-formula>\u0000 loss using pseudo 4-D (i.e., 2.5 Spatial Dimension + 0.5 Temporal Dimension or 2.5SD+0.5TD) patches between kernelized OSEM subset estimates; and 4) a final tuning step which minimizes over-smoothing from the network output within the kernel matrix. Contrast phantom, human [18F]FDG and [11C]RAC data acquired on GE SIGNA PET/MR were used for evaluations. The proposed DL HYPR4D kernel method outperformed the standard HYPR4D kernel method as well as TOF-OSEM and TOF-BSREM (Q.Clear) in terms contrast recovery versus noise. The proposed final tuning reduced the underestimation bias due to over-smoothing within a 4-mm target structure from ~15% to ~2% while maintaining low-noise voxel-level TACs. In addition, the proposed unsupervised DL assisted reconstruction also outperformed the supervised DL version in terms of bias reduction along the TACs and kinetic model fittings.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 1","pages":"20-28"},"PeriodicalIF":4.6,"publicationDate":"2024-08-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142912554","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}
Convergence of the block iterative method in image reconstruction for positron emission tomography (PET) requires careful control of relaxation parameters, which is a challenging task. The automatic determination of relaxation parameters for list-mode reconstructions also remains challenging. Therefore, a different approach would be desirable. In this study, we propose a list-mode maximum-likelihood Dykstra-like splitting PET reconstruction (LM-MLDS) that reduces the limit-cycle amplitude by adding the distance from an initial image as a penalty term into an objective function. LM-MLDS uses a two-step approach because its performance depends on the quality of the initial image. The first step uses a uniform image as the initial image, whereas the second step uses a reconstructed image after one main iteration as the initial image. In a simulation study, LM-MLDS provided a better tradeoff curve between noise and contrast than the other methods. In a clinical study, LM-MLDS removed the false hotspots at the edge of the axial field of view and improved the image quality of slices covering the top of the head to the cerebellum. List-mode proximal splitting reconstruction is useful not only for optimizing nondifferential functions but also for mitigating the limit-cycle phenomenon in block iterative methods.
{"title":"List-Mode PET Image Reconstruction Using Dykstra-Like Splitting","authors":"Kibo Ote;Fumio Hashimoto;Yuya Onishi;Yasuomi Ouchi","doi":"10.1109/TRPMS.2024.3441526","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3441526","url":null,"abstract":"Convergence of the block iterative method in image reconstruction for positron emission tomography (PET) requires careful control of relaxation parameters, which is a challenging task. The automatic determination of relaxation parameters for list-mode reconstructions also remains challenging. Therefore, a different approach would be desirable. In this study, we propose a list-mode maximum-likelihood Dykstra-like splitting PET reconstruction (LM-MLDS) that reduces the limit-cycle amplitude by adding the distance from an initial image as a penalty term into an objective function. LM-MLDS uses a two-step approach because its performance depends on the quality of the initial image. The first step uses a uniform image as the initial image, whereas the second step uses a reconstructed image after one main iteration as the initial image. In a simulation study, LM-MLDS provided a better tradeoff curve between noise and contrast than the other methods. In a clinical study, LM-MLDS removed the false hotspots at the edge of the axial field of view and improved the image quality of slices covering the top of the head to the cerebellum. List-mode proximal splitting reconstruction is useful not only for optimizing nondifferential functions but also for mitigating the limit-cycle phenomenon in block iterative methods.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"9 1","pages":"29-39"},"PeriodicalIF":4.6,"publicationDate":"2024-08-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142912497","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：