Skeletal muscle has a robust regeneration ability that is impaired by severe injury,disease,and aging.resulting in a decline in skeletal muscle function.Therefore,improving skeletal muscle regeneration is a key challe...Skeletal muscle has a robust regeneration ability that is impaired by severe injury,disease,and aging.resulting in a decline in skeletal muscle function.Therefore,improving skeletal muscle regeneration is a key challenge in treating skeletal muscle-related disorders.Owing to their significant role in tissue regeneration,implantation of M2 macrophages(M2MФ)has great potential for improving skeletal muscle regeneration.Here,we present a short-wave infrared(SWIR)fluorescence imaging technique to obtain more in vivo information for an in-depth evaluation of the skeletal muscle regeneration effect after M2MФtransplantation.SWIR fluorescence imaging was employed to track implanted M2MФin the injured skeletal muscle of mouse models.It is found that the implanted M2MФaccumulated at the injury site for two weeks.Then,SWIR fluorescence imaging of blood vessels showed that M2MФimplantation could improve the relative perfusion ratio on day 5(1.09±0.09 vs 0.85±0.05;p=0.01)and day 9(1.38±0.16 vs 0.95±0.03;p=0.01)post-injury,as well as augment the degree of skeletal muscle regencration on day 13 post-injury.Finally,multiple linear regression analyses determined that post-injury time and relative perfusion ratio could be used as predictive indicators to evaluate skeletal muscle regeneration.These results provide more in vivo details about M2MФin skeletal muscle regeneration and confirm that M2MФcould promote angiogenesis and improve the degree of skeletal muscle repair,which will guide the research and development of M2MФimplantation to improve skeletal muscle regeneration.展开更多
We introduce Quafu-Qcover,an open-source cloud-based software package developed for solving combinatorial optimization problems using quantum simulators and hardware backends.Quafu-Qcover provides a standardized and c...We introduce Quafu-Qcover,an open-source cloud-based software package developed for solving combinatorial optimization problems using quantum simulators and hardware backends.Quafu-Qcover provides a standardized and comprehensive workflow that utilizes the quantum approximate optimization algorithm(QAOA).It facilitates the automatic conversion of the original problem into a quadratic unconstrained binary optimization(QUBO)model and its corresponding Ising model,which can be subsequently transformed into a weight graph.The core of Qcover relies on a graph decomposition-based classical algorithm,which efficiently derives the optimal parameters for the shallow QAOA circuit.Quafu-Qcover incorporates a dedicated compiler capable of translating QAOA circuits into physical quantum circuits that can be executed on Quafu cloud quantum computers.Compared to a general-purpose compiler,our compiler demonstrates the ability to generate shorter circuit depths,while also exhibiting superior speed performance.Additionally,the Qcover compiler has the capability to dynamically create a library of qubits coupling substructures in real-time,utilizing the most recent calibration data from the superconducting quantum devices.This ensures that computational tasks can be assigned to connected physical qubits with the highest fidelity.The Quafu-Qcover allows us to retrieve quantum computing sampling results using a task ID at any time,enabling asynchronous processing.Moreover,it incorporates modules for results preprocessing and visualization,facilitating an intuitive display of solutions for combinatorial optimization problems.We hope that Quafu-Qcover can serve as an instructive illustration for how to explore application problems on the Quafu cloud quantum computers.展开更多
With the rapid advancement of quantum computing,hybrid quantum–classical machine learning has shown numerous potential applications at the current stage,with expectations of being achievable in the noisy intermediate...With the rapid advancement of quantum computing,hybrid quantum–classical machine learning has shown numerous potential applications at the current stage,with expectations of being achievable in the noisy intermediate-scale quantum(NISQ)era.Quantum reinforcement learning,as an indispensable study,has recently demonstrated its ability to solve standard benchmark environments with formally provable theoretical advantages over classical counterparts.However,despite the progress of quantum processors and the emergence of quantum computing clouds,implementing quantum reinforcement learning algorithms utilizing parameterized quantum circuits(PQCs)on NISQ devices remains infrequent.In this work,we take the first step towards executing benchmark quantum reinforcement problems on real devices equipped with at most 136 qubits on the BAQIS Quafu quantum computing cloud.The experimental results demonstrate that the policy agents can successfully accomplish objectives under modified conditions in both the training and inference phases.Moreover,we design hardware-efficient PQC architectures in the quantum model using a multi-objective evolutionary algorithm and develop a learning algorithm that is adaptable to quantum devices.We hope that the Quafu-RL can be a guiding example to show how to realize machine learning tasks by taking advantage of quantum computers on the quantum cloud platform.展开更多
AIM: To explore hemodynamics and vasoactive substance levels during renal vein congestion that occurs in the anhepatic phase of liver transplantation.METHODS: New Zealand rabbits received ligation of the hepatic pedic...AIM: To explore hemodynamics and vasoactive substance levels during renal vein congestion that occurs in the anhepatic phase of liver transplantation.METHODS: New Zealand rabbits received ligation of the hepatic pedicle, supra-hepatic vena cava and infrahepatic vena cava [anhepatic phase group(APH); n = 8], the renal veins(RVL; n = 8), renal veins and hepatic pedicle [with the inferior vena cava left open)(RVHP; n = 8)], or a sham operation(SOP; n = 8). Hemodynamic parameters(systolic, diastolic, and mean arterial blood pressures) and the levels of serum bradykinin(BK) and angiotensin Ⅱ(ANGII) were measured at baseline(0 min), and 10 min, 20 min, 30 min, and 45 min after the surgery. Correlation analyses were performed to evaluate the associations between hemodynamic parameters and levels of vasoactive substances.RESULTS:All experimental groups(APH,RVL,and RVHP)showed significant decreases in hemodynamic parameters(systolic,diastolic,and mean arterial blood pressures)compared to baseline levels,as well as compared to the SOP controls(P<0.05 for all).In contrast,BK levels were significantly increased compared to baseline in the APH,RVL,and RVHP groups at all time points measured(P<0.05 for all),whereas no change was observed in the SOP controls.There were no significant differences among the experimental groups for any measure at any time point.Further analyses revealed that systolic,diastolic,and mean arterial blood pressures were all negatively correlated with BK levels,and positively correlated with ANGII levels in the APH,RVL,and RVHP groups(P<0.05 for all).CONCLUSION:In the anhepatic phase of orthotopic liver transplantation,renal vein congestion significantly impacts hemodynamic parameters,which correlate with serum BK and ANGII levels.展开更多
In a fractured porous hydrocarbon reservoir,wave velocities and refections depend on frequency and incident angle.A proper description of the frequency dependence of amplitude variations with ofset(AVO)signatures shou...In a fractured porous hydrocarbon reservoir,wave velocities and refections depend on frequency and incident angle.A proper description of the frequency dependence of amplitude variations with ofset(AVO)signatures should allow efects of fracture inflls and attenuation and dispersion of fractured media.The novelty of this study lies in the introduction of an improved approach for the investigation of incident-angle and frequency variations-associated refection responses.The improved AVO modeling method,using a frequency-domain propagator matrix method,is feasible to accurately consider velocity dispersion predicted from frequency-dependent elasticities from a rock physics modeling.And hence,the method is suitable for use in the case of an anisotropic medium with aligned fractures.Additionally,the proposed modeling approach allows the combined contributions of layer thickness,interbedded structure,impedance contrast and interferences to frequency-dependent refection coefcients and,hence,yielding seismograms of a layered model with a dispersive and attenuative reservoir.Our numerical results show bulk modulus of fracture fuid signifcantly afects anisotropic attenuation,hence causing frequencydependent refection abnormalities.These implications indicate the study of amplitude versus angle and frequency(AVAF)variations provides insights for better interpretation of refection anomalies and hydrocarbon identifcation in a layered reservoir with vertical transverse isotropy(VTI)dispersive media.展开更多
Attacks on websites and network servers are among the most critical threats in network security.Network behavior identification is one of the most effective ways to identify malicious network intrusions.Analyzing abno...Attacks on websites and network servers are among the most critical threats in network security.Network behavior identification is one of the most effective ways to identify malicious network intrusions.Analyzing abnormal network traffic patterns and traffic classification based on labeled network traffic data are among the most effective approaches for network behavior identification.Traditional methods for network traffic classification utilize algorithms such as Naive Bayes,Decision Tree and XGBoost.However,network traffic classification,which is required for network behavior identification,generally suffers from the problem of low accuracy even with the recently proposed deep learning models.To improve network traffic classification accuracy thus improving network intrusion detection rate,this paper proposes a new network traffic classification model,called ArcMargin,which incorporates metric learning into a convolutional neural network(CNN)to make the CNN model more discriminative.ArcMargin maps network traffic samples from the same category more closely while samples from different categories are mapped as far apart as possible.The metric learning regularization feature is called additive angular margin loss,and it is embedded in the object function of traditional CNN models.The proposed ArcMargin model is validated with three datasets and is compared with several other related algorithms.According to a set of classification indicators,the ArcMargin model is proofed to have better performances in both network traffic classification tasks and open-set tasks.Moreover,in open-set tasks,the ArcMargin model can cluster unknown data classes that do not exist in the previous training dataset.展开更多
使用10个LoRa设备和示波器在视距(line of sight,LOS)信道、非视距(non line of sight,NLOS)信道、有扰信道下进行了数据采集并构建了数据集。为了解决当输入为一维时序数据时坐标注意力(coordinate attention,CA)只能在时域上做特征增...使用10个LoRa设备和示波器在视距(line of sight,LOS)信道、非视距(non line of sight,NLOS)信道、有扰信道下进行了数据采集并构建了数据集。为了解决当输入为一维时序数据时坐标注意力(coordinate attention,CA)只能在时域上做特征增强,提出一种DCTCA机制,将输入特征图通过离散余弦变换(discrete cosine transform,DCT)由时域转换到频域以增强在频域上的特征,将时域上的特征图与频域上的注意力图融合实现多维度的特征增强。嵌入到由残差网络(residual network,ResNet)和门控循环网络(gated recurrent unit,GRU)级联的DRGNN网络进行射频指纹特征提取并完成识别。实验结果表明,在有扰信道下网络模型识别准确率可达79.2%,明显优于CNN1D的67.7%和LSTM的45.8%.。通过对比消融实验证明了DCTCA机制的有效性。展开更多
The administration time is a critical but long-neglected point in cell therapy based on macrophages because the incorrect time of macrophage administration could result in diverse outcomes regarding the same macrophag...The administration time is a critical but long-neglected point in cell therapy based on macrophages because the incorrect time of macrophage administration could result in diverse outcomes regarding the same macrophage therapy.In this work,the second near-infrared(NIR-II)fluorescence imaging in vivo tracking of M2 macrophages during a pro-healing therapy in the mice model of rotator cuff injury revealed that the behavior of administrated macrophages was influenced by the timing of their administration.The delayed cell therapy(DCT)group had a longer retention time of injected M2 macrophages in the repairing tissue than that in the immediate cell therapy(ICT)group.Both Keller-Segel model and histological analysis further demonstrated that DCT altered the chemotaxis of M2 macrophages and improved the healing outcome of the repaired structure in comparison with ICT.Our results offer a possible explanation of previous conflicting results on reparative cell therapy and provoke reconsideration of the timing of these therapies.展开更多
Clinical target volume (CTV) delineation is crucial for tumor control and normal tissue protection. This study aimed to define the locoregional extension patterns of nasopharyngeal carcinoma (NPC) and to improve CTV d...Clinical target volume (CTV) delineation is crucial for tumor control and normal tissue protection. This study aimed to define the locoregional extension patterns of nasopharyngeal carcinoma (NPC) and to improve CTV delineation. Magnetic resonance imaging scans of 2366 newly diagnosed NPC patients were reviewed. According to incidence rates of tumor invasion, the anatomic sites surrounding the nasopharynx were classified into high-risk (>30%), medium-risk (5%-30%), and low-risk (<5%) groups. The lymph node (LN) level was determined according to the Radiation Therapy Oncology Group guidelines, which were further categorized into the upper neck (retropharyngeal region and level Ⅱ), middle neck (levels Ⅲ and Va), and lower neck (levels Ⅳ and Vb and the supraclavicular fossa). The high-risk anatomic sites were adjacent to the nasopharynx, whereas those at medium- or low-risk were separated from the nasopharynx. If the high-risk anatomic sites were involved, the rates of tumor invasion into the adjacent medium-risk sites increased; if not, the rates were significantly lower (P < 0.01). Among the 1920 (81.1%) patients with positive LN, the incidence rates of LN metastasis in the upper, middle, and lower neck were 99.6% , 30.2%, and 7.2%, respectively, and skip metastasis happened in only 1.2% of patients. In the 929 patients who had unilateral upper neck involvement, the rates of contralateral middle neck and lower neck involvement were 1.8% and 0.4%, respectively. Thus, local disease spreads stepwise from proximal sites to distal sites, and LN metastasis spreads from the upper neck to the lower neck. Individualized CTV delineation for NPC may be feasible.展开更多
基金supported by Shanghai Sailing Program(22YF1438700)National Key Research and Development Program of China(2021YFA1201303)+5 种基金National Natural Science Foundation of China(82172511,81972121,81972129,82072521,82011530023,and 82111530200)Sanming Project of Medicine in Shenzhen(SZSM201612078)the Introduction Project of Clinical Medicine Expert Team for Suzhou(SZYJTD201714)Shanghai Talent Development Funding Scheme 2020080Shanghai Sailing Program(21YF1404100 and 22YF1405200)Research Project of Shanghai Science and Technology Commission(22DZ2204900)。
文摘Skeletal muscle has a robust regeneration ability that is impaired by severe injury,disease,and aging.resulting in a decline in skeletal muscle function.Therefore,improving skeletal muscle regeneration is a key challenge in treating skeletal muscle-related disorders.Owing to their significant role in tissue regeneration,implantation of M2 macrophages(M2MФ)has great potential for improving skeletal muscle regeneration.Here,we present a short-wave infrared(SWIR)fluorescence imaging technique to obtain more in vivo information for an in-depth evaluation of the skeletal muscle regeneration effect after M2MФtransplantation.SWIR fluorescence imaging was employed to track implanted M2MФin the injured skeletal muscle of mouse models.It is found that the implanted M2MФaccumulated at the injury site for two weeks.Then,SWIR fluorescence imaging of blood vessels showed that M2MФimplantation could improve the relative perfusion ratio on day 5(1.09±0.09 vs 0.85±0.05;p=0.01)and day 9(1.38±0.16 vs 0.95±0.03;p=0.01)post-injury,as well as augment the degree of skeletal muscle regencration on day 13 post-injury.Finally,multiple linear regression analyses determined that post-injury time and relative perfusion ratio could be used as predictive indicators to evaluate skeletal muscle regeneration.These results provide more in vivo details about M2MФin skeletal muscle regeneration and confirm that M2MФcould promote angiogenesis and improve the degree of skeletal muscle repair,which will guide the research and development of M2MФimplantation to improve skeletal muscle regeneration.
基金supported by the National Natural Science Foundation of China(Grant No.92365206)the support of the China Postdoctoral Science Foundation(Certificate Number:2023M740272)+1 种基金supported by the National Natural Science Foundation of China(Grant No.12247168)China Postdoctoral Science Foundation(Certificate Number:2022TQ0036)。
文摘We introduce Quafu-Qcover,an open-source cloud-based software package developed for solving combinatorial optimization problems using quantum simulators and hardware backends.Quafu-Qcover provides a standardized and comprehensive workflow that utilizes the quantum approximate optimization algorithm(QAOA).It facilitates the automatic conversion of the original problem into a quadratic unconstrained binary optimization(QUBO)model and its corresponding Ising model,which can be subsequently transformed into a weight graph.The core of Qcover relies on a graph decomposition-based classical algorithm,which efficiently derives the optimal parameters for the shallow QAOA circuit.Quafu-Qcover incorporates a dedicated compiler capable of translating QAOA circuits into physical quantum circuits that can be executed on Quafu cloud quantum computers.Compared to a general-purpose compiler,our compiler demonstrates the ability to generate shorter circuit depths,while also exhibiting superior speed performance.Additionally,the Qcover compiler has the capability to dynamically create a library of qubits coupling substructures in real-time,utilizing the most recent calibration data from the superconducting quantum devices.This ensures that computational tasks can be assigned to connected physical qubits with the highest fidelity.The Quafu-Qcover allows us to retrieve quantum computing sampling results using a task ID at any time,enabling asynchronous processing.Moreover,it incorporates modules for results preprocessing and visualization,facilitating an intuitive display of solutions for combinatorial optimization problems.We hope that Quafu-Qcover can serve as an instructive illustration for how to explore application problems on the Quafu cloud quantum computers.
基金supported by the Beijing Academy of Quantum Information Sciencessupported by the National Natural Science Foundation of China(Grant No.92365206)+2 种基金the support of the China Postdoctoral Science Foundation(Certificate Number:2023M740272)supported by the National Natural Science Foundation of China(Grant No.12247168)China Postdoctoral Science Foundation(Certificate Number:2022TQ0036)。
文摘With the rapid advancement of quantum computing,hybrid quantum–classical machine learning has shown numerous potential applications at the current stage,with expectations of being achievable in the noisy intermediate-scale quantum(NISQ)era.Quantum reinforcement learning,as an indispensable study,has recently demonstrated its ability to solve standard benchmark environments with formally provable theoretical advantages over classical counterparts.However,despite the progress of quantum processors and the emergence of quantum computing clouds,implementing quantum reinforcement learning algorithms utilizing parameterized quantum circuits(PQCs)on NISQ devices remains infrequent.In this work,we take the first step towards executing benchmark quantum reinforcement problems on real devices equipped with at most 136 qubits on the BAQIS Quafu quantum computing cloud.The experimental results demonstrate that the policy agents can successfully accomplish objectives under modified conditions in both the training and inference phases.Moreover,we design hardware-efficient PQC architectures in the quantum model using a multi-objective evolutionary algorithm and develop a learning algorithm that is adaptable to quantum devices.We hope that the Quafu-RL can be a guiding example to show how to realize machine learning tasks by taking advantage of quantum computers on the quantum cloud platform.
基金Supported by Natural Science Foundation of Gansu Province,China,No.3ZS051-A25-104Clinical Medicine Research Special Funds of Chinese Medical Association,China,No.14040360573
文摘AIM: To explore hemodynamics and vasoactive substance levels during renal vein congestion that occurs in the anhepatic phase of liver transplantation.METHODS: New Zealand rabbits received ligation of the hepatic pedicle, supra-hepatic vena cava and infrahepatic vena cava [anhepatic phase group(APH); n = 8], the renal veins(RVL; n = 8), renal veins and hepatic pedicle [with the inferior vena cava left open)(RVHP; n = 8)], or a sham operation(SOP; n = 8). Hemodynamic parameters(systolic, diastolic, and mean arterial blood pressures) and the levels of serum bradykinin(BK) and angiotensin Ⅱ(ANGII) were measured at baseline(0 min), and 10 min, 20 min, 30 min, and 45 min after the surgery. Correlation analyses were performed to evaluate the associations between hemodynamic parameters and levels of vasoactive substances.RESULTS:All experimental groups(APH,RVL,and RVHP)showed significant decreases in hemodynamic parameters(systolic,diastolic,and mean arterial blood pressures)compared to baseline levels,as well as compared to the SOP controls(P<0.05 for all).In contrast,BK levels were significantly increased compared to baseline in the APH,RVL,and RVHP groups at all time points measured(P<0.05 for all),whereas no change was observed in the SOP controls.There were no significant differences among the experimental groups for any measure at any time point.Further analyses revealed that systolic,diastolic,and mean arterial blood pressures were all negatively correlated with BK levels,and positively correlated with ANGII levels in the APH,RVL,and RVHP groups(P<0.05 for all).CONCLUSION:In the anhepatic phase of orthotopic liver transplantation,renal vein congestion significantly impacts hemodynamic parameters,which correlate with serum BK and ANGII levels.
基金This work was financially supported by the Science Foundation of China University of Petroleum(Beijing)(2462020YXZZ008)the National Natural Science Foundation of China(41804104,41930425,U19B6003-04-03,41774143)+2 种基金the National Key R&D Program of China(2018YFA0702504)the PetroChina Innovation Foundation(2018D-5007-0303)the Science Foundation of SINOPEC Key Laboratory of Geophysics(33550006-20-ZC0699-0001).
文摘In a fractured porous hydrocarbon reservoir,wave velocities and refections depend on frequency and incident angle.A proper description of the frequency dependence of amplitude variations with ofset(AVO)signatures should allow efects of fracture inflls and attenuation and dispersion of fractured media.The novelty of this study lies in the introduction of an improved approach for the investigation of incident-angle and frequency variations-associated refection responses.The improved AVO modeling method,using a frequency-domain propagator matrix method,is feasible to accurately consider velocity dispersion predicted from frequency-dependent elasticities from a rock physics modeling.And hence,the method is suitable for use in the case of an anisotropic medium with aligned fractures.Additionally,the proposed modeling approach allows the combined contributions of layer thickness,interbedded structure,impedance contrast and interferences to frequency-dependent refection coefcients and,hence,yielding seismograms of a layered model with a dispersive and attenuative reservoir.Our numerical results show bulk modulus of fracture fuid signifcantly afects anisotropic attenuation,hence causing frequencydependent refection abnormalities.These implications indicate the study of amplitude versus angle and frequency(AVAF)variations provides insights for better interpretation of refection anomalies and hydrocarbon identifcation in a layered reservoir with vertical transverse isotropy(VTI)dispersive media.
基金This work was supported by the National Natural Science Foundation of China(61871046).
文摘Attacks on websites and network servers are among the most critical threats in network security.Network behavior identification is one of the most effective ways to identify malicious network intrusions.Analyzing abnormal network traffic patterns and traffic classification based on labeled network traffic data are among the most effective approaches for network behavior identification.Traditional methods for network traffic classification utilize algorithms such as Naive Bayes,Decision Tree and XGBoost.However,network traffic classification,which is required for network behavior identification,generally suffers from the problem of low accuracy even with the recently proposed deep learning models.To improve network traffic classification accuracy thus improving network intrusion detection rate,this paper proposes a new network traffic classification model,called ArcMargin,which incorporates metric learning into a convolutional neural network(CNN)to make the CNN model more discriminative.ArcMargin maps network traffic samples from the same category more closely while samples from different categories are mapped as far apart as possible.The metric learning regularization feature is called additive angular margin loss,and it is embedded in the object function of traditional CNN models.The proposed ArcMargin model is validated with three datasets and is compared with several other related algorithms.According to a set of classification indicators,the ArcMargin model is proofed to have better performances in both network traffic classification tasks and open-set tasks.Moreover,in open-set tasks,the ArcMargin model can cluster unknown data classes that do not exist in the previous training dataset.
文摘使用10个LoRa设备和示波器在视距(line of sight,LOS)信道、非视距(non line of sight,NLOS)信道、有扰信道下进行了数据采集并构建了数据集。为了解决当输入为一维时序数据时坐标注意力(coordinate attention,CA)只能在时域上做特征增强,提出一种DCTCA机制,将输入特征图通过离散余弦变换(discrete cosine transform,DCT)由时域转换到频域以增强在频域上的特征,将时域上的特征图与频域上的注意力图融合实现多维度的特征增强。嵌入到由残差网络(residual network,ResNet)和门控循环网络(gated recurrent unit,GRU)级联的DRGNN网络进行射频指纹特征提取并完成识别。实验结果表明,在有扰信道下网络模型识别准确率可达79.2%,明显优于CNN1D的67.7%和LSTM的45.8%.。通过对比消融实验证明了DCTCA机制的有效性。
基金the approval of ethics by Ethics Committee of Fudan University(No.202208005Z)supported by the National Natural Science Foundation of China(Nos.81972129,82072521,82111530200)+1 种基金Shanghai Talent Development Funding Scheme(No.2020080)Shanghai Committee of Science and Technology(Nos.22DZ2204900,23ZR1445700)。
文摘The administration time is a critical but long-neglected point in cell therapy based on macrophages because the incorrect time of macrophage administration could result in diverse outcomes regarding the same macrophage therapy.In this work,the second near-infrared(NIR-II)fluorescence imaging in vivo tracking of M2 macrophages during a pro-healing therapy in the mice model of rotator cuff injury revealed that the behavior of administrated macrophages was influenced by the timing of their administration.The delayed cell therapy(DCT)group had a longer retention time of injected M2 macrophages in the repairing tissue than that in the immediate cell therapy(ICT)group.Both Keller-Segel model and histological analysis further demonstrated that DCT altered the chemotaxis of M2 macrophages and improved the healing outcome of the repaired structure in comparison with ICT.Our results offer a possible explanation of previous conflicting results on reparative cell therapy and provoke reconsideration of the timing of these therapies.
文摘Clinical target volume (CTV) delineation is crucial for tumor control and normal tissue protection. This study aimed to define the locoregional extension patterns of nasopharyngeal carcinoma (NPC) and to improve CTV delineation. Magnetic resonance imaging scans of 2366 newly diagnosed NPC patients were reviewed. According to incidence rates of tumor invasion, the anatomic sites surrounding the nasopharynx were classified into high-risk (>30%), medium-risk (5%-30%), and low-risk (<5%) groups. The lymph node (LN) level was determined according to the Radiation Therapy Oncology Group guidelines, which were further categorized into the upper neck (retropharyngeal region and level Ⅱ), middle neck (levels Ⅲ and Va), and lower neck (levels Ⅳ and Vb and the supraclavicular fossa). The high-risk anatomic sites were adjacent to the nasopharynx, whereas those at medium- or low-risk were separated from the nasopharynx. If the high-risk anatomic sites were involved, the rates of tumor invasion into the adjacent medium-risk sites increased; if not, the rates were significantly lower (P < 0.01). Among the 1920 (81.1%) patients with positive LN, the incidence rates of LN metastasis in the upper, middle, and lower neck were 99.6% , 30.2%, and 7.2%, respectively, and skip metastasis happened in only 1.2% of patients. In the 929 patients who had unilateral upper neck involvement, the rates of contralateral middle neck and lower neck involvement were 1.8% and 0.4%, respectively. Thus, local disease spreads stepwise from proximal sites to distal sites, and LN metastasis spreads from the upper neck to the lower neck. Individualized CTV delineation for NPC may be feasible.