Positron emission tomography (PET) is a widely used functional imaging technique in clinic. Compared to single-tracer PET, dual-tracer dynamic PET allows two sequences of different nuclear pharmaceuticals in one scan, revealing richer physiological information. However, dynamically separating the mixed signals in dual-tracer PET is challenging due to identical energy ~511 keV in gamma photon pairs from both tracers. We propose a method for dynamic PET dual-tracer separation based on invertible neural networks (DTS-INNs). This network enables the forward and backward process simultaneously. Therefore, producing the mixed image sequences from the separation results through backward process may impose extra constraints for optimizing the network. The loss is composed of two components corresponding to the forward and backward propagation processes, which results in more accurate gradient computations and more stable gradient propagation with cycle consistency. We assess our model’s performance using simulated and real data, comparing it with several reputable dual-tracer separation methods. The results of DTS-INN outperform counterparts with lower-mean square error, higher-structural similarity, and peak signal to noise ratio. Additionally, it exhibits robustness against noise levels, phantoms, tracer combinations, and scanning protocols, offering a dependable solution for dual-tracer PET image separation.
正电子发射断层扫描（PET）是一种广泛应用于临床的功能成像技术。与单示踪剂正电子发射计算机断层成像相比，双示踪剂动态正电子发射计算机断层成像允许在一次扫描中使用两种不同的核药物序列，从而揭示更丰富的生理信息。然而，由于两种示踪剂的伽马光子对能量（约 511 keV）相同，动态分离双示踪剂 PET 中的混合信号具有挑战性。我们提出了一种基于可逆神经网络（DTS-INN）的动态 PET 双示踪剂分离方法。该网络可同时进行前向和后向处理。因此，通过后向处理根据分离结果生成混合图像序列可能会对网络的优化造成额外的限制。损耗由对应于前向和后向传播过程的两个部分组成，这使得梯度计算更加精确，梯度传播更加稳定，并具有周期一致性。我们使用模拟数据和真实数据评估了我们模型的性能，并将其与几种著名的双踪分离方法进行了比较。DTS-INN 的结果以更低的均方误差、更高的结构相似性和峰值信噪比优于同类方法。此外，DTS-INN 对噪音水平、模型、示踪剂组合和扫描方案都有很好的适应性，为双示踪剂 PET 图像分离提供了可靠的解决方案。
{"title":"Temporal Image Sequence Separation in Dual-Tracer Dynamic PET With an Invertible Network","authors":"Chuanfu Sun;Bin Huang;Jie Sun;Yangfan Ni;Huafeng Liu;Qian Xia;Qiegen Liu;Wentao Zhu","doi":"10.1109/TRPMS.2024.3407120","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3407120","url":null,"abstract":"Positron emission tomography (PET) is a widely used functional imaging technique in clinic. Compared to single-tracer PET, dual-tracer dynamic PET allows two sequences of different nuclear pharmaceuticals in one scan, revealing richer physiological information. However, dynamically separating the mixed signals in dual-tracer PET is challenging due to identical energy ~511 keV in gamma photon pairs from both tracers. We propose a method for dynamic PET dual-tracer separation based on invertible neural networks (DTS-INNs). This network enables the forward and backward process simultaneously. Therefore, producing the mixed image sequences from the separation results through backward process may impose extra constraints for optimizing the network. The loss is composed of two components corresponding to the forward and backward propagation processes, which results in more accurate gradient computations and more stable gradient propagation with cycle consistency. We assess our model’s performance using simulated and real data, comparing it with several reputable dual-tracer separation methods. The results of DTS-INN outperform counterparts with lower-mean square error, higher-structural similarity, and peak signal to noise ratio. Additionally, it exhibits robustness against noise levels, phantoms, tracer combinations, and scanning protocols, offering a dependable solution for dual-tracer PET image separation.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"774-787"},"PeriodicalIF":4.6,"publicationDate":"2024-03-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10542421","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143785","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 : 2024-03-29DOI: 10.1109/TRPMS.2024.3382990
Ming Niu;Zhonghua Kuang;Xiaohui Wang;Ning Ren;Ziru Sang;Tao Sun;Zheng Liu;Zhanli Hu;Zheng Gu;Yongfeng Yang
The main focus of this work is to compare different timing measurement methods of individual silicon photomultiplier (SiPM) arrays and dual-ended readout PET detectors. Two lutetium yttrium oxyorthosilicate (LYSO) crystal arrays with $3.10times 3.10times 20$ -${mathrm { mm}}^{3}$ crystals, enhanced specular reflector (ESR), and barium sulfate (BaSO4) reflector and one LYSO crystal array with $1.88times 1.88times 20$ -${mathrm { mm}}^{3}$ crystals and $rm BaSO_{4}$ reflector with dual-ended read out by $8times 8$ SiPM arrays of $3times 3$ -${mathrm { mm}}^{2}$ active pixel area were measured. Signals of the SiPM arrays were processed individually using 64 channel PETsys TOFPET2 application specific integrated circuits designed for time-of-flight PET applications. For the SiPM arrays, an energy square-weighted average timing method using the timings of the fastest 2 SiPM pixels was found to provide the best-coincidence timing resolutions (CTRs). For the dual-ended readout detectors, the method of using the energy-weighted average timings of the two SiPM arrays provided the best CTR of 234 ps for the detector using $3.10times 3.10times 20$ -${mathrm { mm}}^{3}$ crystals and ESR reflector, 239 ps for the detector using $3.10times 3.10times 20$ -${mathrm { mm}}^{3}$ crystals and $rm BaSO_{4}$ reflector, and 275 ps for the detector using $1.88times 1.88times 20$ -${mathrm { mm}}^{3}$ crystals and $rm BaSO_{4}$ reflector for an energy window of 410–610 keV. The dual-ended readout detectors developed in this work provide better CTRs than those of single-ended readout detectors and a high-3-D position resolution which can be used in the future to develop whole-body PET scanners to simultaneously achieve uniform high-spatial resolution, high sensitivity and high-timing resolution.
{"title":"Comparison of Timing Measurement Methods of Dual-Ended Readout Scintillator Array PET Detectors","authors":"Ming Niu;Zhonghua Kuang;Xiaohui Wang;Ning Ren;Ziru Sang;Tao Sun;Zheng Liu;Zhanli Hu;Zheng Gu;Yongfeng Yang","doi":"10.1109/TRPMS.2024.3382990","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3382990","url":null,"abstract":"The main focus of this work is to compare different timing measurement methods of individual silicon photomultiplier (SiPM) arrays and dual-ended readout PET detectors. Two lutetium yttrium oxyorthosilicate (LYSO) crystal arrays with \u0000<inline-formula> <tex-math>$3.10times 3.10times 20$ </tex-math></inline-formula>\u0000-\u0000<inline-formula> <tex-math>${mathrm { mm}}^{3}$ </tex-math></inline-formula>\u0000 crystals, enhanced specular reflector (ESR), and barium sulfate (BaSO4) reflector and one LYSO crystal array with \u0000<inline-formula> <tex-math>$1.88times 1.88times 20$ </tex-math></inline-formula>\u0000-\u0000<inline-formula> <tex-math>${mathrm { mm}}^{3}$ </tex-math></inline-formula>\u0000 crystals and \u0000<inline-formula> <tex-math>$rm BaSO_{4}$ </tex-math></inline-formula>\u0000 reflector with dual-ended read out by \u0000<inline-formula> <tex-math>$8times 8$ </tex-math></inline-formula>\u0000 SiPM arrays of \u0000<inline-formula> <tex-math>$3times 3$ </tex-math></inline-formula>\u0000-\u0000<inline-formula> <tex-math>${mathrm { mm}}^{2}$ </tex-math></inline-formula>\u0000 active pixel area were measured. Signals of the SiPM arrays were processed individually using 64 channel PETsys TOFPET2 application specific integrated circuits designed for time-of-flight PET applications. For the SiPM arrays, an energy square-weighted average timing method using the timings of the fastest 2 SiPM pixels was found to provide the best-coincidence timing resolutions (CTRs). For the dual-ended readout detectors, the method of using the energy-weighted average timings of the two SiPM arrays provided the best CTR of 234 ps for the detector using \u0000<inline-formula> <tex-math>$3.10times 3.10times 20$ </tex-math></inline-formula>\u0000-\u0000<inline-formula> <tex-math>${mathrm { mm}}^{3}$ </tex-math></inline-formula>\u0000 crystals and ESR reflector, 239 ps for the detector using \u0000<inline-formula> <tex-math>$3.10times 3.10times 20$ </tex-math></inline-formula>\u0000-\u0000<inline-formula> <tex-math>${mathrm { mm}}^{3}$ </tex-math></inline-formula>\u0000 crystals and \u0000<inline-formula> <tex-math>$rm BaSO_{4}$ </tex-math></inline-formula>\u0000 reflector, and 275 ps for the detector using \u0000<inline-formula> <tex-math>$1.88times 1.88times 20$ </tex-math></inline-formula>\u0000-\u0000<inline-formula> <tex-math>${mathrm { mm}}^{3}$ </tex-math></inline-formula>\u0000 crystals and \u0000<inline-formula> <tex-math>$rm BaSO_{4}$ </tex-math></inline-formula>\u0000 reflector for an energy window of 410–610 keV. The dual-ended readout detectors developed in this work provide better CTRs than those of single-ended readout detectors and a high-3-D position resolution which can be used in the future to develop whole-body PET scanners to simultaneously achieve uniform high-spatial resolution, high sensitivity and high-timing resolution.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 6","pages":"607-617"},"PeriodicalIF":4.6,"publicationDate":"2024-03-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10485386","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141500325","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 : 2024-03-28DOI: 10.1109/TRPMS.2024.3381849
Mohamed Kassar;Milos Drobnjakovic;Gabriele Birindelli;Song Xue;Andrei Gafita;Thomas Wendler;Ali Afshar-Oromieh;Nassir Navab;Wolfgang A. Weber;Matthias Eiber;Sibylle Ziegler;Axel Rominger;Kuangyu Shi
Pretherapy dosimetry prediction is a prerequisite for treatment planning and personalized optimization of the emerging radiopharmaceutical therapy (RPT). Physiologically based pharmacokinetic (PBPK) model, describing the intrinsic pharmacokinetics of radiopharmaceuticals, have been proposed for pretherapy prediction of dosimetry. However, it is restricted with organwise prediction and the customization based on pretherapy measurements is still challenging. On the other side, artificial intelligence (AI) has demonstrated the potential in pretherapy dosimetry prediction. Nevertheless, it is still challenging for pure data-driven model to achieve voxelwise prediction due to huge gap between the pretherapy imaging and post-therapy dosimetry. This study aims to integrate the PBPK model into deep learning for voxelwise pretherapy dosimetry prediction. A conditional generative adversarial network (cGAN) integrated with the PBPK model as regularization was developed. For proof of concept, 120 virtual patients with 68Ga-PSMA-11 PET imaging and 177Lu-PSMA-I&T dosimetry were generated using realistic in silico simulations. In kidneys, spleen, liver and salivary glands, the proposed method achieved better accuracy and dose volume histogram than pure deep learning. The preliminary results confirmed that the proposed PBPK-adapted deep learning can improve the pretherapy voxelwise dosimetry prediction and may provide a practical solution to support treatment planning of heterogeneous dose distribution for personalized RPT.
{"title":"PBPK-Adapted Deep Learning for Pretherapy Prediction of Voxelwise Dosimetry: In-Silico Proof of Concept","authors":"Mohamed Kassar;Milos Drobnjakovic;Gabriele Birindelli;Song Xue;Andrei Gafita;Thomas Wendler;Ali Afshar-Oromieh;Nassir Navab;Wolfgang A. Weber;Matthias Eiber;Sibylle Ziegler;Axel Rominger;Kuangyu Shi","doi":"10.1109/TRPMS.2024.3381849","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3381849","url":null,"abstract":"Pretherapy dosimetry prediction is a prerequisite for treatment planning and personalized optimization of the emerging radiopharmaceutical therapy (RPT). Physiologically based pharmacokinetic (PBPK) model, describing the intrinsic pharmacokinetics of radiopharmaceuticals, have been proposed for pretherapy prediction of dosimetry. However, it is restricted with organwise prediction and the customization based on pretherapy measurements is still challenging. On the other side, artificial intelligence (AI) has demonstrated the potential in pretherapy dosimetry prediction. Nevertheless, it is still challenging for pure data-driven model to achieve voxelwise prediction due to huge gap between the pretherapy imaging and post-therapy dosimetry. This study aims to integrate the PBPK model into deep learning for voxelwise pretherapy dosimetry prediction. A conditional generative adversarial network (cGAN) integrated with the PBPK model as regularization was developed. For proof of concept, 120 virtual patients with 68Ga-PSMA-11 PET imaging and 177Lu-PSMA-I&T dosimetry were generated using realistic in silico simulations. In kidneys, spleen, liver and salivary glands, the proposed method achieved better accuracy and dose volume histogram than pure deep learning. The preliminary results confirmed that the proposed PBPK-adapted deep learning can improve the pretherapy voxelwise dosimetry prediction and may provide a practical solution to support treatment planning of heterogeneous dose distribution for personalized RPT.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 6","pages":"646-654"},"PeriodicalIF":4.6,"publicationDate":"2024-03-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10481675","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141500252","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 : 2024-03-26DOI: 10.1109/TRPMS.2024.3381865
Jiahao Xie;Haibo Wang;Simon R. Cherry;Junwei Du
Almost all high spatial resolution positron emission tomography (PET) detectors based on pixelated scintillator arrays utilize crystal arrays with smaller pitches than photodetector arrays, leading to challenges in resolving edge crystals. To address this issue, this article introduces a novel multi-resolution silicon photomultiplier (SiPM) array design aimed at decreasing the number of readout channels required while maintaining the crystal resolvability of the detector, especially for edge crystals. The performance of a pseudo $9times9$ multi-resolution SiPM array, consisting of $6.47times6.47$ mm 2, $6.47times3.07$ mm 2, and $3.07times3.07$ mm2 SiPMs, was compared to those of a pseudo $8times8$ SiPM array with a 6.8-mm pitch, and a $16times16$ SiPM array with a 3.4-mm pitch using a $36times36$ LYSO array with a pitch of 1.5 mm. The large-size pseudo SiPMs were implemented by digitally grouping multiple $3.07times3.07$ mm2 SiPMs. The flood histograms show that the edge crystal resolvability of the pseudo $9times9$ multi-resolution SiPM array is comparable to that of the $16times16$ SiPM array and is significantly better than that of the $8times8$ SiPM array.
{"title":"PET Detectors Based on Multi-Resolution SiPM Arrays","authors":"Jiahao Xie;Haibo Wang;Simon R. Cherry;Junwei Du","doi":"10.1109/TRPMS.2024.3381865","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3381865","url":null,"abstract":"Almost all high spatial resolution positron emission tomography (PET) detectors based on pixelated scintillator arrays utilize crystal arrays with smaller pitches than photodetector arrays, leading to challenges in resolving edge crystals. To address this issue, this article introduces a novel multi-resolution silicon photomultiplier (SiPM) array design aimed at decreasing the number of readout channels required while maintaining the crystal resolvability of the detector, especially for edge crystals. The performance of a pseudo \u0000<inline-formula> <tex-math>$9times9$ </tex-math></inline-formula>\u0000 multi-resolution SiPM array, consisting of \u0000<inline-formula> <tex-math>$6.47times6.47$ </tex-math></inline-formula>\u0000 mm 2, \u0000<inline-formula> <tex-math>$6.47times3.07$ </tex-math></inline-formula>\u0000 mm 2, and \u0000<inline-formula> <tex-math>$3.07times3.07$ </tex-math></inline-formula>\u0000 mm2 SiPMs, was compared to those of a pseudo \u0000<inline-formula> <tex-math>$8times8$ </tex-math></inline-formula>\u0000 SiPM array with a 6.8-mm pitch, and a \u0000<inline-formula> <tex-math>$16times16$ </tex-math></inline-formula>\u0000 SiPM array with a 3.4-mm pitch using a \u0000<inline-formula> <tex-math>$36times36$ </tex-math></inline-formula>\u0000 LYSO array with a pitch of 1.5 mm. The large-size pseudo SiPMs were implemented by digitally grouping multiple \u0000<inline-formula> <tex-math>$3.07times3.07$ </tex-math></inline-formula>\u0000 mm2 SiPMs. The flood histograms show that the edge crystal resolvability of the pseudo \u0000<inline-formula> <tex-math>$9times9$ </tex-math></inline-formula>\u0000 multi-resolution SiPM array is comparable to that of the \u0000<inline-formula> <tex-math>$16times16$ </tex-math></inline-formula>\u0000 SiPM array and is significantly better than that of the \u0000<inline-formula> <tex-math>$8times8$ </tex-math></inline-formula>\u0000 SiPM array.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 5","pages":"493-500"},"PeriodicalIF":4.4,"publicationDate":"2024-03-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820283","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 : 2024-03-25DOI: 10.1109/TRPMS.2024.3379215
Md Ashequr Rahman;Zitong Yu;Richard Laforest;Craig K. Abbey;Barry A. Siegel;Abhinav K. Jha
There is an important need for methods to process myocardial perfusion imaging (MPI) single-photon emission computed tomography (SPECT) images acquired at lower-radiation dose and/or acquisition time such that the processed images improve observer performance on the clinical task of detecting perfusion defects compared to low-dose images. To address this need, we build upon concepts from model-observer theory and our understanding of the human visual system to propose a detection task-specific deep-learning-based approach for denoising MPI SPECT images (DEMIST). The approach, while performing denoising, is designed to preserve features that influence observer performance on detection tasks. We objectively evaluated DEMIST on the task of detecting perfusion defects using a retrospective study with anonymized clinical data in patients who underwent MPI studies across two scanners ($N,,=$ 338). The evaluation was performed at low-dose levels of 6.25%, 12.5%, and 25% and using an anthropomorphic channelized Hotelling observer. Performance was quantified using area under the receiver operating characteristics curve (AUC). Images denoised with DEMIST yielded significantly higher AUC compared to corresponding low-dose images and images denoised with a commonly used task-agnostic deep learning-based denoising method. Similar results were observed with stratified analysis based on patient sex and defect type. Additionally, DEMIST improved visual fidelity of the low-dose images as quantified using root mean squared error and structural similarity index metric. A mathematical analysis revealed that DEMIST preserved features that assist in detection tasks while improving the noise properties, resulting in improved observer performance. The results provide strong evidence for further clinical evaluation of DEMIST to denoise low-count images in MPI SPECT.
{"title":"DEMIST: A Deep-Learning-Based Detection-Task-Specific Denoising Approach for Myocardial Perfusion SPECT","authors":"Md Ashequr Rahman;Zitong Yu;Richard Laforest;Craig K. Abbey;Barry A. Siegel;Abhinav K. Jha","doi":"10.1109/TRPMS.2024.3379215","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3379215","url":null,"abstract":"There is an important need for methods to process myocardial perfusion imaging (MPI) single-photon emission computed tomography (SPECT) images acquired at lower-radiation dose and/or acquisition time such that the processed images improve observer performance on the clinical task of detecting perfusion defects compared to low-dose images. To address this need, we build upon concepts from model-observer theory and our understanding of the human visual system to propose a detection task-specific deep-learning-based approach for denoising MPI SPECT images (DEMIST). The approach, while performing denoising, is designed to preserve features that influence observer performance on detection tasks. We objectively evaluated DEMIST on the task of detecting perfusion defects using a retrospective study with anonymized clinical data in patients who underwent MPI studies across two scanners (\u0000<inline-formula> <tex-math>$N,,=$ </tex-math></inline-formula>\u0000 338). The evaluation was performed at low-dose levels of 6.25%, 12.5%, and 25% and using an anthropomorphic channelized Hotelling observer. Performance was quantified using area under the receiver operating characteristics curve (AUC). Images denoised with DEMIST yielded significantly higher AUC compared to corresponding low-dose images and images denoised with a commonly used task-agnostic deep learning-based denoising method. Similar results were observed with stratified analysis based on patient sex and defect type. Additionally, DEMIST improved visual fidelity of the low-dose images as quantified using root mean squared error and structural similarity index metric. A mathematical analysis revealed that DEMIST preserved features that assist in detection tasks while improving the noise properties, resulting in improved observer performance. The results provide strong evidence for further clinical evaluation of DEMIST to denoise low-count images in MPI SPECT.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 4","pages":"439-450"},"PeriodicalIF":4.4,"publicationDate":"2024-03-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10477616","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140342755","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 : 2024-03-22DOI: 10.1109/TRPMS.2024.3403959
Ryosuke Ota;Kibo Ote
Bismuth germanate (BGO) has been receiving attention again because it is a potential scintillator for future time-of-flight positron emission tomography. Owing to its optical properties, BGO emits a relatively large number of Cherenkov photons after 511-keV gamma-ray interactions, which can enable good coincidence time resolution (CTR). Nonetheless, optimally exploiting the Cherenkov emissions can be confounded by scintillation emissions. Thus, we propose a method efficiently emphasizing Cherenkov photon from a detector waveform by deconvolving a single photon response of photodetector. As a proof-of-concept, we perform the deconvolution, and a probability density function (PDF) of BGO was obtained, which is compared to a conventional time correlated single photon counting (TCSPC) method. Furthermore, we investigate if the proposed deconvolution can emphasize a faint Cherenkov signal. Consequently, the PDF obtained by the proposed deconvolution shows a good agreement with that obtained using a conventional TCSPC methods. A CTR obtained using the proposed deconvolution is improved by 12% and 43% in full width at half maximum compared to a voltage-based leading edge discriminator for with and without high-frequency readout electronics, respectively. Thus, the proposed deconvolution method can efficiently emphasize Cherenkov photon by lowering the threshold level and improve the timing performance of BGO-based detectors.
锗酸铋（BGO）再次受到关注，因为它是未来飞行时间正电子发射断层扫描的潜在闪烁体。由于其光学特性，锗酸铋在与 511-keV 伽马射线相互作用后会发出相对较多的切伦科夫光子，从而实现良好的重合时间分辨率（CTR）。然而，最佳利用切伦科夫发射可能会受到闪烁发射的干扰。因此，我们提出了一种方法，通过对光电探测器的单光子响应进行解卷积，从探测器波形中有效地强调切伦科夫光子。作为概念验证，我们进行了解卷积，得到了 BGO 的概率密度函数（PDF），并与传统的时间相关单光子计数（TCSPC）方法进行了比较。此外，我们还研究了拟议的解卷积是否能突出微弱的切伦科夫信号。结果表明，拟议解卷积得到的 PDF 与传统 TCSPC 方法得到的 PDF 非常吻合。与基于电压的前沿鉴别器相比，在有高频读出电子设备和无高频读出电子设备的情况下，利用拟议解卷积法获得的 CTR 在半最大全宽方面分别提高了 12% 和 43%。因此，建议的解卷积方法可以通过降低阈值水平有效地强调切伦科夫光子，并改善基于 BGO 的探测器的计时性能。
{"title":"Emphasizing Cherenkov Photons From Bismuth Germanate by Single Photon Response Deconvolution","authors":"Ryosuke Ota;Kibo Ote","doi":"10.1109/TRPMS.2024.3403959","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3403959","url":null,"abstract":"Bismuth germanate (BGO) has been receiving attention again because it is a potential scintillator for future time-of-flight positron emission tomography. Owing to its optical properties, BGO emits a relatively large number of Cherenkov photons after 511-keV gamma-ray interactions, which can enable good coincidence time resolution (CTR). Nonetheless, optimally exploiting the Cherenkov emissions can be confounded by scintillation emissions. Thus, we propose a method efficiently emphasizing Cherenkov photon from a detector waveform by deconvolving a single photon response of photodetector. As a proof-of-concept, we perform the deconvolution, and a probability density function (PDF) of BGO was obtained, which is compared to a conventional time correlated single photon counting (TCSPC) method. Furthermore, we investigate if the proposed deconvolution can emphasize a faint Cherenkov signal. Consequently, the PDF obtained by the proposed deconvolution shows a good agreement with that obtained using a conventional TCSPC methods. A CTR obtained using the proposed deconvolution is improved by 12% and 43% in full width at half maximum compared to a voltage-based leading edge discriminator for with and without high-frequency readout electronics, respectively. Thus, the proposed deconvolution method can efficiently emphasize Cherenkov photon by lowering the threshold level and improve the timing performance of BGO-based detectors.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 6","pages":"595-606"},"PeriodicalIF":4.6,"publicationDate":"2024-03-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141500389","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 : 2024-03-21DOI: 10.1109/TRPMS.2024.3380090
S. M. A. Sharif;Rizwan Ali Naqvi;Woong-Kee Loh
Medical image denoising is considered among the most challenging vision tasks. Despite the real-world implications, existing denoising methods have notable drawbacks as they often generate visual artifacts when applied to heterogeneous medical images. This study addresses the limitation of the contemporary denoising methods with an artificial intelligence (AI)-driven two-stage learning strategy. The proposed method learns to estimate the residual noise from the noisy images. Later, it incorporates a novel noise attention mechanism to correlate estimated residual noise with noisy inputs to perform denoising in a course-to-refine manner. This study also proposes to leverage a multimodal learning strategy to generalize the denoising among medical image modalities and multiple noise patterns for widespread applications. The practicability of the proposed method has been evaluated with dense experiments. The experimental results demonstrated that the proposed method achieved state-of-the-art performance by significantly outperforming the existing medical image denoising methods in quantitative and qualitative comparisons. Overall, it illustrates a performance gain of 7.64 in peak signal-to-noise ratio (PSNR), 0.1021 in structural similarity index (SSIM), 0.80 in DeltaE $(Delta E)$ , 0.1855 in visual information fidelity pixelwise (VIFP), and 18.54 in mean squared error (MSE) metrics.
{"title":"Two-Stage Deep Denoising With Self-Guided Noise Attention for Multimodal Medical Images","authors":"S. M. A. Sharif;Rizwan Ali Naqvi;Woong-Kee Loh","doi":"10.1109/TRPMS.2024.3380090","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3380090","url":null,"abstract":"Medical image denoising is considered among the most challenging vision tasks. Despite the real-world implications, existing denoising methods have notable drawbacks as they often generate visual artifacts when applied to heterogeneous medical images. This study addresses the limitation of the contemporary denoising methods with an artificial intelligence (AI)-driven two-stage learning strategy. The proposed method learns to estimate the residual noise from the noisy images. Later, it incorporates a novel noise attention mechanism to correlate estimated residual noise with noisy inputs to perform denoising in a course-to-refine manner. This study also proposes to leverage a multimodal learning strategy to generalize the denoising among medical image modalities and multiple noise patterns for widespread applications. The practicability of the proposed method has been evaluated with dense experiments. The experimental results demonstrated that the proposed method achieved state-of-the-art performance by significantly outperforming the existing medical image denoising methods in quantitative and qualitative comparisons. Overall, it illustrates a performance gain of 7.64 in peak signal-to-noise ratio (PSNR), 0.1021 in structural similarity index (SSIM), 0.80 in DeltaE \u0000<inline-formula> <tex-math>$(Delta E)$ </tex-math></inline-formula>\u0000, 0.1855 in visual information fidelity pixelwise (VIFP), and 18.54 in mean squared error (MSE) metrics.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 5","pages":"521-531"},"PeriodicalIF":4.4,"publicationDate":"2024-03-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820221","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}
In the planning phase of radiation therapy, positron emission tomography (PET) images are frequently integrated with computed tomography (CT) and MRI to accurately delineate the target region for treatment. However, obtaining additional CT or magnetic resonance (MR) images solely for localization purposes proves to be financially burdensome, time-intensive, and may increase patient radiation exposure. To alleviate these issues, a deep learning model with dynamic modality translation capabilities is introduced. This approach is achieved through the incorporation of adaptive modality translation layers within the decoder module. The adaptive modality translation layer effectively governs modality transformation by reshaping the data distribution of features extracted by the encoder using switch codes. The model’s performance is assessed on images with reference images using evaluation metrics, such as peak signal-to-noise ratio, structural similarity index measure, and normalized mean square error. For results without reference images, subjective assessments are provided by six nuclear medicine physicians based on clinical interpretations. The proposed model demonstrates impressive performance in transforming nonattenuation corrected PET images into user-specified modalities (attenuation corrected PET, MR, or CT), effectively streamlining the acquisition of supplemental modality images in radiation therapy scenarios.
在放射治疗的计划阶段，正电子发射断层扫描（PET）图像经常与计算机断层扫描（CT）和磁共振成像（MRI）结合使用，以准确划定治疗靶区。然而，仅为定位目的而获取额外的 CT 或磁共振（MR）图像不仅经济负担重、耗时长，还可能增加患者的辐射量。为了缓解这些问题，我们引入了一种具有动态模式转换能力的深度学习模型。这种方法是通过在解码器模块中加入自适应模态转换层来实现的。自适应模态转换层通过重塑编码器使用开关代码提取的特征的数据分布，有效控制模态转换。利用峰值信噪比、结构相似性指数测量和归一化均方误差等评估指标，对带有参考图像的图像进行模型性能评估。对于无参考图像的结果，则由六位核医学医生根据临床解释进行主观评估。所提出的模型在将非衰减校正 PET 图像转换为用户指定模式（衰减校正 PET、MR 或 CT）方面表现出色，有效简化了放射治疗场景中补充模式图像的获取。
{"title":"Medical Multimodal Image Transformation With Modality Code Awareness","authors":"Zhihua Li;Yuxi Jin;Qingneng Li;Zhenxing Huang;Zixiang Chen;Chao Zhou;Na Zhang;Xu Zhang;Wei Fan;Jianmin Yuan;Qiang He;Weiguang Zhang;Dong Liang;Zhanli Hu","doi":"10.1109/TRPMS.2024.3379580","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3379580","url":null,"abstract":"In the planning phase of radiation therapy, positron emission tomography (PET) images are frequently integrated with computed tomography (CT) and MRI to accurately delineate the target region for treatment. However, obtaining additional CT or magnetic resonance (MR) images solely for localization purposes proves to be financially burdensome, time-intensive, and may increase patient radiation exposure. To alleviate these issues, a deep learning model with dynamic modality translation capabilities is introduced. This approach is achieved through the incorporation of adaptive modality translation layers within the decoder module. The adaptive modality translation layer effectively governs modality transformation by reshaping the data distribution of features extracted by the encoder using switch codes. The model’s performance is assessed on images with reference images using evaluation metrics, such as peak signal-to-noise ratio, structural similarity index measure, and normalized mean square error. For results without reference images, subjective assessments are provided by six nuclear medicine physicians based on clinical interpretations. The proposed model demonstrates impressive performance in transforming nonattenuation corrected PET images into user-specified modalities (attenuation corrected PET, MR, or CT), effectively streamlining the acquisition of supplemental modality images in radiation therapy scenarios.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 5","pages":"511-520"},"PeriodicalIF":4.4,"publicationDate":"2024-03-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820224","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 : 2024-03-18DOI: 10.1109/TRPMS.2024.3378421
S. Di Giacomo;M. Ronchi;G. Borghi;D. R. Schaart;M. Carminati;C. Fiorini
Embedding signal processing in the front-end of radiation detectors represents an approach to cope with the growing complexity of nuclear imaging scanners with increasing field of view (i.e., higher number of channels). Machine learning (ML) offers a good compromise between intrinsic image reconstruction performance and computational power. While most hardware accelerators for ML are based on digital circuits and, thus, require the analog-to-digital conversion of all individual signals from photodetectors, an analog approach allows to streamline the pipeline. We present the study of an analog accelerator implementing a neural network (NN) with 42 neurons in a 0.35-${mu }$ m CMOS process node. The specific target is the reconstruction of the position of interaction of gamma-rays in the scintillator crystal of Anger cameras used for PET and SPECT. This chip can be used stand-alone or monolithically integrated within the application specific integrated circuit (ASIC) for the filtering of current signals from arrays of silicon photomultipliers (SiPMs). Computation is performed in charge domain by means of crossbar arrays of programmable capacitor. The architecture of the 64-input ASIC and the training of the NN are presented, discussing the impact of weight quantization on 5 bits. From MATLAB and circuit simulations, consistent with ASIC topology and operations, the NN capabilities were tested using two different datasets, obtained from both simulated data and experimental data, both based on PET detector composed by a monolithic scintillator crystal readout by an $8times 8$ array of SiPMs. Simulations show an achievable spatial resolution better than 2-mm full-width-at-half-maximum with a 10-mm thick crystal, a max. count rate of 200kHz and the energy efficiency per inference is estimated to be of 93.5GOP/J, i.e., competitive with digital counterparts, with an energy consumption of 38nJ per inference and area of 23mm2.
随着视野范围的不断扩大（即通道数增加），核成像扫描仪的复杂性也在不断增加，将信号处理嵌入辐射探测器前端是应对这种情况的一种方法。机器学习（ML）是内在图像重建性能和计算能力之间的良好折衷。虽然大多数用于 ML 的硬件加速器都基于数字电路，因此需要对来自光电探测器的所有单独信号进行模数转换，但模拟方法可以简化流水线。我们介绍了对模拟加速器的研究，该加速器在 0.35- ${mu }$ m CMOS 工艺节点上实现了一个拥有 42 个神经元的神经网络 (NN)。具体目标是重建用于 PET 和 SPECT 的 Anger 相机闪烁晶体中伽马射线相互作用的位置。该芯片可独立使用，也可单片集成在专用集成电路（ASIC）中，用于过滤硅光电倍增管（SiPM）阵列发出的电流信号。计算是通过可编程电容器横条阵列在电荷域中进行的。介绍了 64 输入 ASIC 的结构和 NN 的训练，讨论了权重量化对 5 位的影响。根据 MATLAB 和电路仿真（与 ASIC 拓扑结构和操作一致），使用两个不同的数据集测试了 NN 的能力，这两个数据集均来自模拟数据和实验数据，均基于由单片闪烁晶体组成的 PET 检测器，该检测器由 8 美元/次的 SiPM 阵列读出。模拟结果表明，使用 10 毫米厚的晶体，可实现优于 2 毫米全宽半最大值的空间分辨率，最大计数率为 200kHz，每次推理的能效估计为 93.5GOP/J，即与数字同行相比具有竞争力，每次推理的能耗为 38nJ，面积为 23 平方毫米。
{"title":"Implementing an Integrated Neural Network for Real-Time Position Reconstruction in Emission Tomography With Monolithic Scintillators","authors":"S. Di Giacomo;M. Ronchi;G. Borghi;D. R. Schaart;M. Carminati;C. Fiorini","doi":"10.1109/TRPMS.2024.3378421","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3378421","url":null,"abstract":"Embedding signal processing in the front-end of radiation detectors represents an approach to cope with the growing complexity of nuclear imaging scanners with increasing field of view (i.e., higher number of channels). Machine learning (ML) offers a good compromise between intrinsic image reconstruction performance and computational power. While most hardware accelerators for ML are based on digital circuits and, thus, require the analog-to-digital conversion of all individual signals from photodetectors, an analog approach allows to streamline the pipeline. We present the study of an analog accelerator implementing a neural network (NN) with 42 neurons in a 0.35-\u0000<inline-formula> <tex-math>${mu }$ </tex-math></inline-formula>\u0000m CMOS process node. The specific target is the reconstruction of the position of interaction of gamma-rays in the scintillator crystal of Anger cameras used for PET and SPECT. This chip can be used stand-alone or monolithically integrated within the application specific integrated circuit (ASIC) for the filtering of current signals from arrays of silicon photomultipliers (SiPMs). Computation is performed in charge domain by means of crossbar arrays of programmable capacitor. The architecture of the 64-input ASIC and the training of the NN are presented, discussing the impact of weight quantization on 5 bits. From MATLAB and circuit simulations, consistent with ASIC topology and operations, the NN capabilities were tested using two different datasets, obtained from both simulated data and experimental data, both based on PET detector composed by a monolithic scintillator crystal readout by an \u0000<inline-formula> <tex-math>$8times 8$ </tex-math></inline-formula>\u0000 array of SiPMs. Simulations show an achievable spatial resolution better than 2-mm full-width-at-half-maximum with a 10-mm thick crystal, a max. count rate of 200kHz and the energy efficiency per inference is estimated to be of 93.5GOP/J, i.e., competitive with digital counterparts, with an energy consumption of 38nJ per inference and area of 23mm2.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 5","pages":"501-510"},"PeriodicalIF":4.4,"publicationDate":"2024-03-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820220","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 : 2024-03-17DOI: 10.1109/TRPMS.2024.3402272
Zixiang Chen;Yuxi Jin;Zhenxing Huang;Na Zhang;Kaiyi Liang;Guotao Quan;Dong Liang;Hairong Zheng;Zhanli Hu
Dynamic cerebral perfusion CT (PCT) is an effective imaging technique for the clinical diagnosis and therapy guidance of many kinds of cerebrovascular diseases (CVDs), but the large radiation dose imposed on a patient during repeated CT scans greatly limits its clinical applications. Achieving low-dose PCT imaging with the help of advanced and satisfactory imaging methods is needed. A kinetic-induced voxel-clustering filter (KVCF) is proposed in this work to help process noisy and distorted PCT images acquired from the low-dose CT scan protocols. In this new method, the intrinsic kinetic information of objective PCT images is extracted and effectively utilized to construct an image filter for every PCT frame. The new method is validated using both the simulated and clinical low-dose PCT data, and the peak signal-to-noise ratio (PSNR) and feature similarity (FSIM) are applied for quantitative evaluations of both the dynamic images and the calculated hemodynamic parametric maps. Compared to several existing methods, the proposed KVCF method produces the best qualitative and quantitative imaging effects. With satisfactory performance and high interpretability, KVCF is proven to be effective and implementable in the clinical low-dose PCT imaging tasks.
{"title":"Building a Kinetic Induced Voxel-Clustering Filter (KVCF) for Low-Dose Perfusion CT Imaging","authors":"Zixiang Chen;Yuxi Jin;Zhenxing Huang;Na Zhang;Kaiyi Liang;Guotao Quan;Dong Liang;Hairong Zheng;Zhanli Hu","doi":"10.1109/TRPMS.2024.3402272","DOIUrl":"https://doi.org/10.1109/TRPMS.2024.3402272","url":null,"abstract":"Dynamic cerebral perfusion CT (PCT) is an effective imaging technique for the clinical diagnosis and therapy guidance of many kinds of cerebrovascular diseases (CVDs), but the large radiation dose imposed on a patient during repeated CT scans greatly limits its clinical applications. Achieving low-dose PCT imaging with the help of advanced and satisfactory imaging methods is needed. A kinetic-induced voxel-clustering filter (KVCF) is proposed in this work to help process noisy and distorted PCT images acquired from the low-dose CT scan protocols. In this new method, the intrinsic kinetic information of objective PCT images is extracted and effectively utilized to construct an image filter for every PCT frame. The new method is validated using both the simulated and clinical low-dose PCT data, and the peak signal-to-noise ratio (PSNR) and feature similarity (FSIM) are applied for quantitative evaluations of both the dynamic images and the calculated hemodynamic parametric maps. Compared to several existing methods, the proposed KVCF method produces the best qualitative and quantitative imaging effects. With satisfactory performance and high interpretability, KVCF is proven to be effective and implementable in the clinical low-dose PCT imaging tasks.","PeriodicalId":46807,"journal":{"name":"IEEE Transactions on Radiation and Plasma Medical Sciences","volume":"8 7","pages":"762-773"},"PeriodicalIF":4.6,"publicationDate":"2024-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142143650","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：