两种自动勾画软件对危及器官勾画结果对比分析

目的:比较和分析两种自动勾画软件（AccuContour和DeepViewer）勾画危及器官的精确度,以此评估它们在不同肿瘤放射治疗中的适用程度和优越性。方法:回顾性选取中科院合肥肿瘤医院肿瘤患者60例,其中鼻咽癌、肺癌、乳腺癌、宫颈癌各15例,由同一个物理师在患者CT图像上手动勾画危及器官,再分别用两种自动勾画软件进行勾画。以手动勾画结果为标准,分别计算两种软件勾画结果的戴斯相似性系数（DSC）和绝对体积差（ΔV）,并对两种勾画结果的差异进行配对t检验,比较两种软件勾画结果。结果:AccuContour软件和DeepViewer软件勾画结果的总体DSC分别为0.90±0.11和0.87±0.14（t=-5.029, P<0.05）,总体ΔV分别为（13.23±18.77）和（29.89±45.27） cm3（t=7.344, P<0.05）。在20个危及器官中,AccuContour软件勾画结果的所有DSC均大于0.7,其中最大DSC为脑（0.99±0.00）,最小DSC为右眼晶状体（0.71±0.11）;DeepViewer软件勾画的结果有18个器官DSC大于等于0.7,其中,最大DSC为肺（0.98±0.00）,最小DSC为右侧股骨头（0.63±0.18）。AccuContour软件勾画的13个器官的ΔV均小于DeepViewer勾画结果。结论:两种软件整体勾画效果均比较好,对于体积较大的危及器官,勾画效果要优于体积较小的器官,AccuContour软件勾画效果优于DeepViewer软件。     

基于剂量预测和自动勾画技术的PET/CT器官内照射剂量率快速评估方法

目的：实现一种基于深度学习的剂量预测和自动勾画技术的正电子发射断层成像（PET）/CT检查器官内照射剂量率的快速评估方法。方法：首先基于患者特定时刻的PET/CT图像,使用蒙特卡罗程序GATE进行内照射剂量率计算,获得每个患者的剂量率分布图。随后,基于U-Net构建深度神经网络,将患者的CT和PET图像作为输入,GATE计算的剂量率图作为金标准进行训练。训练后的深度学习模型能够根据患者的CT和PET图像预测对应的剂量率分布。同时,使用勾画软件DeepViewer对患者CT图像中的器官和组织进行自动勾画,结合预测得到的剂量率分布结果计算相应器官和组织的吸收剂量率。使用50名患者的PET/CT数据,其中10份用于测试,其余40份进行4折交叉训练,每次使用30份用于训练,10份用于验证。将测试集结果与GATE和GPU蒙特卡罗工具ARCHER-NM进行对比。结果：在自动勾画软件DeepViewer勾画的24个器官中,绝大部分器官的深度学习预测剂量率与GATE计算结果偏差在±10%以内。其中大脑、心脏、肝脏、左肺、右肺的平均偏差分别为3.3%、1.1%、1.0%、-1.1%、0.0%,与GATE...     

TCP/NTCP生物模型在胸中上段食管癌调强放疗中的应用

目的:探究TCP/NTCP生物模型在胸中上段食管癌放疗计划优化中的应用及剂量学特点。方法:回顾性分析47例胸中上段食管癌患者,为每位患者制定基于剂量体积（DV）限制的IMRT计划,在DV计划基础上添加对计划靶区（PTV）的TCP（限值90%、95%）生物模型以及危及器官的NTCP（限值10%、5%）模型优化,依次生成[PlanTCP90%]、[PlanTCP95%]、[PlanNTCP10%]、[PlanNTCP5%]4组计划。从剂量学及生物学参数方面评估计划间差异。结果:[PlanTCP90%]、[PlanTCP95%]相比于DV计划,靶区剂量参数均提高（P<0.05）,其中[Dmean]、[D2%]、[D98%]分别提高（2.1%、9.8%）、（1.9%、9.8%）和（1.7%、9.3%）,CI分别降低5%、20%,HI相近,TCP值分别提高2%、7%;危及器官受照剂量均有不同程度提高,[PlanTCP95%]增加更显著（P<0.05）。经TCP优化后的计划仅[PlanTCP90%]满足临床要求;[PlanNTCP5%]与DV计划相比,靶区、心脏相关剂量参数间差异没有统计学意义（P>0.05）,但双肺[Dmean]、[V5 Gy]、[V10 Gy]、[V20 Gy]、NTCP值降低4.4%、1.6%、2.6%、6.2%、0.52%（P<0.05）。[PlanNTCP10%]与DV计划相比,PTV及危及器官相关剂量学参数间差异没有统计学意义（P>0.05）。结论:TCP/NTCP生物学优化可以使靶区及危及器官剂量更符合生物学要求,建议在胸中上段食管癌放疗DV计划优化后引入TCP/NTCP的评估,进而进行选择性的靶向深入优化。     

基于深度学习的危及器官自动勾画软件系统DeepViewer在放疗中的应用及评估

目的:将一款基于深度学习的危及器官自动勾画软件系统DeepViewer应用于临床,实现自动勾画肿瘤患者治疗计划中危及器官的功能。方法:DeepViewer使用改进后的全卷积神经网络U-Net来实现自动勾画患者CT扫描部位所包含的危及器官,并使用Dice相似性系数（DSC）对比分析这22种危及器官自动勾画与手动勾画的差异。结果:11种危及器官DSC平均值在0.9以上,5种危及器官DSC平均值为0.8~0.9,5种器官DSC平均值为0.7~0.8,视交叉DSC平均值最低,为0.676。总体结果表明DeepViewer系统能够较准确地自动勾画出危及器官,特别是左、右肺、膀胱、脑干等器官,已基本满足临床需求。结论:DeepViewer软件系统可以实现放疗肿瘤患者危及器官的自动勾画,准确性较高。同时,DeepViewer系统勾画完毕后,可以通过网络系统自动传输RTStructure DICOM3.0文件,无需其他操作,能极大地提高临床医生工作效率,降低治疗计划流程中的勾画总时间。     

基于人工智能技术的危及器官自动勾画在胸部肿瘤中的应用

目的:评估基于人工智能技术的自动勾画软件勾画胸部危及器官轮廓的几何学精度,为临床应用提供依据。方法:选择30例胸部肿瘤患者的CT图像,分别使用基于人工智能技术的自动勾画软件勾画和医师手动勾画胸部危及器官。采用Hausdorff距离、形状相似性指数及Jaccard系数这3个指标评价自动勾画与手动勾画危及器官的几何学一致性。结果:在肺、心脏和脊髓的Hausdorff距离中,最大为右肺的（22.31±4.50） mm,最小为脊髓的（3.17±0.80） mm。危及器官的形状相似性指数值均≥0.91。Jaccard系数中左肺和右肺的均值≥0.95,脊髓的为0.84±0.02,心脏的略低为0.83±0.04。结论:基于人工智能技术的危及器官自动勾画软件对于胸部危及器官勾画能够达到较高的准确性和精度,可以满足临床工作。 【关键词】胸部肿瘤;人工智能;危及器官;自动勾画;放射治疗     

宫颈癌放疗中计算网格尺寸对物理剂量和生物剂量的影响

目的:定量分析剂量计算网格尺寸（DCGS）对宫颈癌放疗中物理剂量和生物剂量的影响。方法:选取Pinnacle3治疗计划系统中宫颈癌的治疗方案12例,取默认值DCGS=4.0 mm的计算网格,优化调整宫颈癌治疗方案,再改变DCGS（1.0～7.0 mm）,重新计算靶区和危及器官（OAR）的剂量,探讨靶区和OAR的物理剂量和生物剂量随DCGS的变化情况。结果:靶区和OAR的物理剂量随DCGS的变大而减小,在体积剂量直方图上表现出曲线整体向低剂量区平移。除左右股骨头外,靶区的肿瘤控制概率（TCP）和OAR的正常组织并发症概率（NTCP）也随DCGS增大而缓慢降低。PGTVnd的TCP下降率约为0.7%/mm,PTV的TCP下降率约为0.6%/mm,而膀胱和直肠的NTCP下降速度相对较快,膀胱NTCP下降率最大值为15.0%,直肠NTCP下降率最大值为13.5%。结论:宫颈癌放疗中物理剂量和生物剂量随DCGS变大而减少,靶区和OAR的物理剂量在体积剂量直方图上表现出整体向低剂量区平移,这种变化趋势会诱导研究者低估靶区的TCP及OAR的NTCP。     

TomoTherapy在全身放疗方面的应用

目的：本文报道一例TomoTherapy全身照射病例,介绍其调强计划方案,并分析其计划及验证结果。方法：由定位CT获取病人全身CT图像,在CMS医生工作站进行靶区和重要器官的勾画,传至TomoTherapy计划工作站进行计划设计。用ArcCheck三维半导体矩阵对计划实施验证,计划验证通过后在螺旋断层放疗设备上完成对病人的全身照射。结果：85%的靶区接受10Gy的处方剂量,靶区适形度CI和均匀指数HI分别为0.76和1.16。重要器官限制剂量达到临床要求,左晶体的最大剂量为3.4Gy,右晶体的最大剂量为4.3Gy,肺的平均剂量为5.5Gy,V5为45%,左肾的平均剂量为5Gy,右肾的平均剂量为5.7Gy。结论：TomoTherapy可以完成全身的照射,肺的剂量可以降得更低,减少毒副反应。为全身放疗提供了一种安全的方法。     

基于U-net卷积神经网络的宫颈癌临床靶区和危及器官自动勾画的研究

目的:基于U-net卷积神经网络的深度学习方法,探讨宫颈癌放疗临床靶区和危及器官自动勾画的可行性。方法:利用U-net卷积神经网络模型搭建的端到端自动分割框架,以100例已进行IMRT治疗的宫颈癌患者CT及组织结构信息为研究对象,并随机选取其中的10例作为测试集。勾画的对象包括临床靶区（CTV）、膀胱、直肠和左、右股骨头5个部分,比较手动和自动勾画的戴斯相似性系数（DSC）和豪斯多夫距离（HD）以评估自动勾画模型的准确性。结果:4种危及器官自动勾画的DSC值都在0.833以上,平均值是0.898;HD值均在8.3 mm以内,平均值为5.3 mm;临床靶区DSC值是0.860,HD值为13.9 mm。结论:基于U-net卷积神经网络建立的自动勾画模型能较为准确地实现宫颈癌临床靶区和危及器官的自动勾画,临床应用中可大幅提高医生的工作效率及勾画的一致性。     

放疗影像自动分割效果评估中几何参数与剂量学参数之间的关联性

目的:通过分析感兴趣区域（ROI）的几何参数与剂量学参数之间的关联性,探讨放疗影像自动分割效果评估时联合使用几何参数与剂量学参数的必要性。方法:利用卷积神经网络构建的分割模型对18例宫颈癌术后患者的危及器官与靶区进行自动分割,把自动分割结果与医生手动勾画结果进行比较,用于评估的几何参数包括基于体积/面积的Dice相似性系数、相对体积差与基于距离的几何参数:最大Hausdorff距离、95% Hausdorff距离、质心差,剂量学参数包括针对危及器官的平均剂量差、针对靶区的ΔD95和ΔD98。采用线性回归方法研究不同分割方式下ROI几何学参数与剂量学参数间的关系,并使用Spearman相关性分析获得几何参数间的相关性及医生勾画与自动分割间剂量学的相关性。结果:所有ROI的几何参数与剂量学参数间的关系均较弱（63.3%的R2<0.4）且不稳定;同时几何参数间的相关系数|r|不超过0.625,互为弱相关或不相关。结论:在对放疗领域的图像分割结果进行评估时,应该同时考虑到几何参数与剂量学参数。选择几何参数时,应联合基于面积/体积的评估方式与基于距离的评估方式。     

CT造影剂对脑动静脉畸形VMAT计划剂量分布的影响

目的:探讨CT造影剂对脑动静脉畸形VMAT计划剂量分布的影响。方法:选取15例脑动静脉畸形患者为研究对象，在相同体位下行平扫和增强CT扫描。在增强CT图像上勾画靶区和危及器官，通过图像配准将上述结构复制至平扫CT图像。在VARIAN Eclipse 13.6计划系统中完成增强CT图像的VMAT计划设计（5条弧），并将此计划复制至平扫CT图像，不进行通量优化，重新计算剂量分布。记录两组计划的靶区、危及器官以及感兴趣区域的CT值，靶区的D2%、D98%、Dmean、适形度指数、均匀性指数以及梯度跌落指数，危及器官Dmax以及正常组织受照射体积V2 Gy、V10 Gy、V12 Gy，并采用非参数配对Wilcoxon检验分析两组数据间的差异。结果:在靶区和感兴趣区域处，增强图像的CT值显著高于平扫图像（P=0.001）；而在其他组织处，两组图像的CT值比较均无统计学差异（P>0.05）。两组计划的剂量学参数差异小于2%，且无统计学差异（P>0.05）。结论:CT造影剂对脑动静脉畸形VMAT计划剂量分布的影响较小，临床上可忽略。     

Comparison of whole-body phantom designs to estimate organ equivalent neutron doses for secondary cancer risk assessment in proton therapy

Secondary neutron fluence created during proton therapy can be a significant source of radiation exposure in organs distant from the treatment site, especially in pediatric patients. Various published studies have used computational phantoms to estimate neutron equivalent doses in proton therapy. In these simulations, whole-body patient representations were applied considering either generic whole-body phantoms or generic age- and gender-dependent phantoms. No studies to date have reported using patient-specific geometry information. The purpose of this study was to estimate the effects of patient–phantom matching when using computational pediatric phantoms. To achieve this goal, three sets of phantoms, including different ages and genders, were compared to the patients' whole-body CT. These sets consisted of pediatric age specific reference, age-adjusted reference and anatomically sculpted phantoms. The neutron equivalent dose for a subset of out-of-field organs was calculated using the GEANT4 Monte Carlo toolkit, where proton fields were used to irradiate the cranium and the spine of all phantoms and the CT-segmented patient models. The maximum neutron equivalent dose per treatment absorbed dose was calculated and found to be on the order of 0 to 5 mSv Gy(-1). The relative dose difference between each phantom and their respective CT-segmented patient model for most organs showed a dependence on how close the phantom and patient heights were matched. The weight matching was found to have much smaller impact on the dose accuracy except for very heavy patients. Analysis of relative dose difference with respect to height difference suggested that phantom sculpting has a positive effect in terms of dose accuracy as long as the patient is close to the 50th percentile height and weight. Otherwise, the benefit of sculpting was masked by inherent uncertainties, i.e. variations in organ shapes, sizes and locations.Other sources of uncertainty included errors associated with beam positioning, neutron weighting factor definition and organ segmentation. This work demonstrated the importance of hybrid phantom height matching for more accurate organ dose calculation in proton therapy and the potential limitations of reference phantoms released by regulatory bodies for radiation therapy applications.     

Monte Carlo simulation estimates of neutron doses to critical organs of a patient undergoing 18 MV x-ray LINAC-based radiotherapy

Absorbed photoneutron dose to patients undergoing 18 MV x-ray therapy was studied using Monte Carlo simulations based on the MCNPX code. Two separate transport simulations were conducted, one for the photoneutron contribution and another for neutron capture gamma rays. The phantom model used was of a female patient receiving a four-field pelvic box treatment. Photoneutron doses were determinate to be higher for organs and tissues located inside the treatment field, especially those closest to the patient's skin. The maximum organ equivalent dose per x-ray treatment dose achieved within each treatment port was 719 microSv/Gy to the rectum (180 degrees field), 190 microSv/Gy to the intestine wall (0 degrees field), 51 microSv/Gy to the colon wall (90 degrees field), and 45 microSv/Gy to the skin (270 degrees field). The maximum neutron equivalent dose per x-ray treatment dose received by organs outside the treatment field was 65 microSv/Gy to the skin in the antero-posterior field. A mean value of 5 +/- 2 microSv/Gy was obtained for organs distant from the treatment field. Distant organ neutron equivalent doses are all of the same order of magnitude and constitute a good estimate of deep organ neutron equivalent doses. Using the risk assessment method of the ICRP-60 report, the greatest likelihood of fatal secondary cancer for a 70 Gy dose is estimated to be 0.02% for the pelvic postero-anterior field, the rectum being the organ representing the maximum contribution of 0.011%.     

Limitations of a convolution method for modeling geometric uncertainties in radiation therapy: the radiobiological dose-per-fraction effect

The convolution method can be used to model the effect of random geometric uncertainties into planned dose distributions used in radiation treatment planning. This is effectively done by linearly adding infinitesimally small doses, each with a particular geometric offset, over an assumed infinite number of fractions. However, this process inherently ignores the radiobiological dose-per-fraction effect since only the summed physical dose distribution is generated. The resultant potential error on predicted radiobiological outcome [quantified in this work with tumor control probability (TCP), equivalent uniform dose (EUD), normal tissue complication probability (NTCP), and generalized equivalent uniform dose (gEUD)] has yet to be thoroughly quantified. In this work, the results of a Monte Carlo simulation of geometric displacements are compared to those of the convolution method for random geometric uncertainties of 0, 1, 2, 3, 4, and 5 mm (standard deviation). The alpha/betaCTV ratios of 0.8, 1.5, 3, 5, and 10 Gy are used to represent the range of radiation responses for different tumors, whereas a single alpha/betaOAR ratio of 3 Gy is used to represent all the organs at risk (OAR). The analysis is performed on a four-field prostate treatment plan of 18 MV x rays. The fraction numbers are varied from 1-50, with isoeffective adjustments of the corresponding dose-per-fractions to maintain a constant tumor control, using the linear-quadratic cell survival model. The average differences in TCP and EUD of the target, and in NTCP and gEUD of the OAR calculated from the convolution and Monte Carlo methods reduced asymptotically as the total fraction number increased, with the differences reaching negligible levels beyond the treatment fraction number of > or =20. The convolution method generally overestimates the radiobiological indices, as compared to the Monte Carlo method, for the target volume, and underestimates those for the OAR. These effects are interconnected and attributed to assuming an infinite number of fractions inherent in the implementation of the convolution technique, irrespective of the uniqueness of each treatment schedule. Based on the fraction numbers analyzed (1-50), and the range of fraction numbers normally used clinically (> or =20), the convolution method can be used safely to estimate the effects of random geometric uncertainties on prostate treatment radiobiological outcomes, for both the target and the OAR. Although the results of this study is likely to apply to other clinical sites and treatment techniques other than the four-field, further validation similar to those done in this study may be necessary prior to clinical implementation.     

Lyman-Kutcher-Burman NTCP model parameters for radiation pneumonitis and xerostomia based on combined analysis of published clinical data

Knowledge of accurate parameter estimates is essential for incorporating normal tissue complication probability (NTCP) models into biologically based treatment planning. The purpose of this work is to derive parameter estimates for the Lyman-Kutcher-Burman (LKB) NTCP model using a combined analysis of multi-institutional toxicity data for the lung (radiation pneumonitis) and parotid gland (xerostomia). A series of published clinical datasets describing dose response for radiation pneumonitis (RP) and xerostomia were identified for this analysis. The data support the notion of large volume effect for the lung and parotid gland with the estimates of the n parameter being close to unity. Assuming that n = 1, the m and TD(50) parameters of the LKB model were estimated by the maximum likelihood method from plots of complication rate as a function of mean organ dose. Ninety five percent confidence intervals for parameter estimates were obtained by the profile likelihood method. If daily fractions other than 2 Gy had been used in a published report, mean organ doses were converted to 2 Gy/fraction-equivalent doses using the linear-quadratic (LQ) formula with alpha/beta = 3 Gy. The following parameter estimates were obtained for the endpoint of symptomatic RP when the lung is considered a paired organ: m = 0.41 (95% CI 0.38, 0.45) and TD(50) = 29.9 Gy (95% CI 28.2, 31.8). When RP incidence was evaluated as a function of dose to the ipsilateral lung rather than total lung, estimates were m = 0.35 (95% CI 0.29, 0.43) and TD(50) = 37.6 Gy (95% CI 34.6, 41.4). For xerostomia expressed as reduction in stimulated salivary flow below 25% within six months after radiotherapy, the following values were obtained: m = 0.53 (95% CI 0.45, 0.65) and TD(50) = 31.4 Gy (95% CI 29.1, 34.0). Although a large number of parameter estimates for different NTCP models and critical structures exist and continue to appear in the literature, it is hard to justify the use of any single parameter set obtained at a selected institution for the purposes of biologically based treatment planning. Our expectation is that the proposed model parameters based on cumulative experience at various institutions are more representative of the overall practice of radiation therapy than any single-institution data, and could be more readily incorporated into clinical use.     

A new formula for normal tissue complication probability (NTCP) as a function of equivalent uniform dose (EUD)

To facilitate the use of biological outcome modeling for treatment planning, an exponential function is introduced as a simpler equivalent to the Lyman formula for calculating normal tissue complication probability (NTCP). The single parameter of the exponential function is chosen to reproduce the Lyman calculation to within approximately 0.3%, and thus enable easy conversion of data contained in empirical fits of Lyman parameters for organs at risk (OARs). Organ parameters for the new formula are given in terms of Lyman model m and TD(50), and conversely m and TD(50) are expressed in terms of the parameters of the new equation. The role of the Lyman volume-effect parameter n is unchanged from its role in the Lyman model. For a non-homogeneously irradiated OAR, an equation relates d(ref), n, v(eff) and the Niemierko equivalent uniform dose (EUD), where d(ref) and v(eff) are the reference dose and effective fractional volume of the Kutcher-Burman reduction algorithm (i.e. the LKB model). It follows in the LKB model that uniform EUD irradiation of an OAR results in the same NTCP as the original non-homogeneous distribution. The NTCP equation is therefore represented as a function of EUD. The inverse equation expresses EUD as a function of NTCP and is used to generate a table of EUD versus normal tissue complication probability for the Emami-Burman parameter fits as well as for OAR parameter sets from more recent data.     

Estimating radiation doses from multidetector CT using Monte Carlo simulations: effects of different size voxelized patient models on magnitudes of organ and effective dose

The purpose of this work is to examine the effects of patient size on radiation dose from CT scans. To perform these investigations, we used Monte Carlo simulation methods with detailed models of both patients and multidetector computed tomography (MDCT) scanners. A family of three-dimensional, voxelized patient models previously developed and validated by the GSF was implemented as input files using the Monte Carlo code MCNPX. These patient models represent a range of patient sizes and ages (8 weeks to 48 years) and have all radiosensitive organs previously identified and segmented, allowing the estimation of dose to any individual organ and calculation of patient effective dose. To estimate radiation dose, every voxel in each patient model was assigned both a specific organ index number and an elemental composition and mass density. Simulated CT scans of each voxelized patient model were performed using a previously developed MDCT source model that includes scanner specific spectra, including bowtie filter, scanner geometry and helical source path. The scan simulations in this work include a whole-body scan protocol and a thoracic CT scan protocol, each performed with fixed tube current. The whole-body scan simulation yielded a predictable decrease in effective dose as a function of increasing patient weight. Results from analysis of individual organs demonstrated similar trends, but with some individual variations. A comparison with a conventional dose estimation method using the ImPACT spreadsheet yielded an effective dose of 0.14 mSv mAs(-1) for the whole-body scan. This result is lower than the simulations on the voxelized model designated 'Irene' (0.15 mSv mAs(-1)) and higher than the models 'Donna' and 'Golem' (0.12 mSv mAs(-1)). For the thoracic scan protocol, the ImPACT spreadsheet estimates an effective dose of 0.037 mSv mAs(-1), which falls between the calculated values for Irene (0.042 mSv mAs(-1)) and Donna (0.031 mSv mAs(-1)) and is higher relative to Golem (0.025 mSv mAs(-1)). This work demonstrates the ability to estimate both individual organ and effective doses from any arbitrary CT scan protocol on individual patient-based models and to provide estimates of the effect of patient size on these dose metrics.