基于体外信号的呼吸运动跟踪模型的研究

发布时间：2018-04-09 08:20

本文选题：放射治疗　切入点：呼吸跟踪　出处：《南方医科大学》2012年硕士论文

【摘要】：放射治疗是治疗肿瘤最重要的手段之一,其根本目的在于使肿瘤靶区接受尽可能大剂量的照射,同时周围正常组织接受的剂量尽可能小或者免受照射。随着技术的发展,各种精确放疗技术相继出现并在临床得到广泛的应用,放疗患者的生存率和生活质量得到稳步提升。目前,在放疗过程中,仍然存在着诸多的不确定性,如患者分次间的摆位误差,肿瘤的体积和位置随着治疗的进行发生变化,以及在治疗中患者存在着各种非自主运动,尤其是呼吸运动,会导致胸腹部靶区发生较大的位移。呼吸运动对放疗的影响贯穿着整个治疗过程,包括影响计划图像的采集、影响放疗计划的设计和计划的精确执行等。目前常规的用于处理放疗中呼吸运动的方法包括：运动包含法、压迫式浅呼吸法、屏气法和呼吸门控法等,但这些方法均存在不足,不能很好的解决呼吸运动的带来的问题。实时跟踪法是目前处理呼吸运动的最佳方法,通过跟踪设备实时获取患者的靶区位置信息,然后再将这些信息反馈给射束调整装置,使高能射线始终对准肿瘤靶区,实现对肿瘤的精确治疗。要实现实时跟踪治疗,核心问题之一就是要实现运动靶区准确实时的跟踪。目前应用最广泛的跟踪方法有三种：通过X线成像跟踪植入靶区或者靶区附近的金属标记物、通过X线成像跟踪与靶区同步运动器官和通过光学测量装置跟踪患者体表标记物的运动。使用X线进行跟踪可以准确的获取体内靶区的运动信息,但患者将接受大量额外剂量的照射,且植入标记物的过程是有创的；使用光学法能实时的获取患者的体外运动信息,操作便利且对患者完全无创,但由于体内运动和体外运动之间的关系不恒定,仅通过体外跟踪很难实现对体内靶区的准确跟踪。目前认为最可行的方法是将体内运动测量法和体外运动测量法进行结合,充分利用二者的优势,实现对运动靶区准确实时的跟踪。目前在临床上有着广泛应用的Cyberknife治疗系统的同步呼吸跟踪系统(Synchrony系统)就是基于这一理念。Synchrony系统通过一对正交的X线成像系统跟踪植入体内的金属标记物来获取体内靶区的运动数据；通过红外定位装置获取患者体外的运动信息；在治疗中实时获取体外运动数据,并通过相关性模型推算得到靶区的位置信息；然后通过预测算法提前预知靶区的运动,最后将模型的结果传递给治疗系统用于调整治疗。但Synchrony系统同样存在一些不足,如需通过有创的方法植入标记物、相关性模型和预测模型的误差比较大等。以Synchrony系统的跟踪模型为基础,我们提出了一种基于体外运动信号的呼吸运动跟踪模型,其模型架构与Synchrony系统类似,但各个模块的具体实现存在很大的区别。在模型中,体外运动数据通过NDI公司的POLARIS红外定位系统来采集。本文以NDI提供的通信接口函数为基础,实现了一套实用的体外呼吸运动测量系统,主要功能包括运动数据的采集、显示和记录,在视场的三视图中显示标记物的位置,实时计算呼吸运动参数,通过预测算法对呼吸运动进行预测,显示实时运动曲线、预测曲线和预测误差等。在实验中,将置于患者体表的红外反射标记物的运动数据作为体外呼吸运动数据。体内运动数据通过数字模拟定位机进行采集,以横膈膜顶部的运动信息作为体内运动数据。在透视模式下,通过摄像头记录膈顶的运动过程,再将视频转换为数字图像,通过目标跟踪算法自动在图像中得到膈顶的位置信息。本文实现了三种目标跟踪算法：二维最小绝对差累加和算法(MAD算法),最多邻近点距离算法(MCD算法)和互信息算法(MI算法),结果表明,三种算法均能有效地对运动目标进行跟踪,其中MI算法的准确性和鲁棒性最好,并针对MI算法的匹配速度过慢的问题,采用了一种等步长搜索法对搜索过程进行加速。在治疗过程中,采用的运动跟踪策略是体内低频采样、体外高频采样,从体外运动推算体内运动,因此要求在治疗开始时建立体内运动和体外运动的相关性模型,在治疗时通过跟踪体外运动来获知靶区的位置信息。而且从运动跟踪设备开始跟踪到治疗设备做好调整之间存在着系统延迟,包括数据获取时间、计算处理时间、数据传输及机械延迟时间等,总延迟时间可以达到几百毫秒,处理这一问题最有效的方法是通过预测模型提前预知靶区的位置信息。对于相关性模型,将体外的运动数据作为输入,靶区的运动估计数据作为输出；而对于预测模型,是将当前值作为输入,未来值作为输出,两种模型的本质很相似,故可以使用相同的函数形式,然后依照不同模型给定对应的输入和输出,求出模型对应的参数,即可分别构建出两种模型。但是由于呼吸运动本身很不规则,同时放疗对模型的准确性和实时性要求非常高,利用传统的建模方法很难满足要求,本研究提出应用非参数回归法构建相关性模型和预测模型。本文采集了11名受试者的呼吸运动数据,然后分别使用非参数回归模型、自回归模型和BP神经网络模型进行预测,并与无预测时的结果进行比较。同时针对预测过程中出现的“异常状态”,提出了一种改进的非参数回归预测方法。最后将预测算法集成在测量系统中,以验证预测算法实时测量中的有效性。经测试表明,在不同的预测长度下,非参数回归法能够准确实时的对呼吸运动进行预测,改进的方法则能大幅减小呼吸运动中“异常状态”的预测误差。在预测长度为0.6s时,11组数据在无预测,自回归模型、BP神经网络、非参数回归和改进非参数回归法的归一化均方误差均值分别为0.85,0.54,0.52、0.44和0.4,且与测量系统结合后,算法同样能实时准确的进行预测。 Synchrony系统中所使用的相关性模型为混合多项式模型,其结构简单,但是模型的误差较大,在实验中,我们建立了基于非参数回归法的相关性模型,并且与线性模型、双二次多项式模型和神经网络模型的结果进行比较。使用7组体内-体外同步运动数据进行了验证,其中体内运动数据为通过3D超声获取的肝脏内血管的运动数据,体外数据为通过光学测量系统获得的体表标记物的运动数据。经计算,线性、双二次多项式、神经网络和非参数回归四种相关性模型的归一化均方误差均值分别为0.35、0.32、0.30、0.19,因此基于非参数回归的相关性模型误差远小于其他三种模型,并且模型构建方便,计算的实时性好。通过光学设备跟踪体外运动时,可以同时跟踪多个标记物的运动信息,采样点越多,所包含的体外运动数据就越多,此时模型也会更复杂。在研究中,本文建立了基于非参数回归的多体外-体内运动相关性模型,使用1个、2个、3个体外标记物时模型的归一化均方误差均值分别为0.185、0.136、0.126,由此可知,模型包含的体外标记物越多,模型的误差越小,但模型的误差和体外标记物的组合之间并不存在确定的关系。通过比较不同的算法建立的内外运动相关性模型、运动预测模型和多体外-体内相关性模型,可知,非参数回归法在呼吸运动的建模中具有鲁棒性强、准确度高、实时性好的优点,能够满足实时跟踪放疗的要求。并且随着跟踪的进行,模型的历史数据库不断的扩充,模型的准确性会不断提高。在文中最后对所做的工作做了小结,同时说明了文中存在的一些不足并且对今后的工作做了一些展望。
[Abstract]:Radiotherapy is one of the most important means of cancer treatment, its fundamental purpose is to make the tumor target accept as large as possible dose, while the surrounding normal tissue dose as small as possible or from radiation. With the development of technology, all have the accurate radiotherapy technology and widely used in clinical, patients with radiotherapy the survival rate and life quality steadily.
At present, in the process of radiotherapy, there are still many uncertainties, such as the set-up errors were divided between the tumor size and location as the treatment of changes in the treatment of patients, and there are all kinds of non autonomous motion, especially respiratory movement, will lead to large displacement of chest and abdomen target area effect of radiotherapy on the respiratory movement. Throughout the course of treatment, including the effect of program image acquisition, influence design and planning radiotherapy precision. Methods include conventional radiotherapy for the treatment of respiratory movement in the present: exercise includes pressure method, shallow breathing, breath holding and respiratory gating, but these methods have shortcomings, solve the problems caused by respiratory motion is not very good. The real-time tracking method is the best method with respiratory motion, the target location tracking device of real-time access to patient The information is placed, and then the information is fed back to the beam adjustment device, so that the high-energy rays are always aligned with the target area of the tumor to achieve the precise treatment of the tumor.
In order to realize the real-time tracking of treatment, one of the key issues is to achieve accurate real-time tracking of moving targets. There are three kinds of currently used tracking method widely: tracking metal markers near implanted into the target area or target area by X-ray imaging, X-ray imaging tracking and target organ of synchronous motion and motion tracking in patients with skin marker the optical measuring device. The use of X-ray motion tracking information can be extracted in the target area, but a large number of patients will receive an extra dose of irradiation, the process and the implantation of markers is invasive; using optical method can obtain real-time information of patients in vitro, convenient operation and is harmless to patients. But because of the relationship between in vivo and in vitro sports movement is not constant, only by in vitro tracking is difficult to achieve accurate tracking of internal target. At present the most feasible method is to use. In the motion measurement and motion measurement method combined with in vitro, make full use of the advantages of the two, to achieve accurate real-time tracking of moving targets. The current clinical respiratory synchronization with Cyberknife treatment system is widely applied in the tracking system (Synchrony system) motion data is metal markers to track the in vivo x-ray imaging system of the the concept of.Synchrony system through a pair of orthogonal based to obtain the internal target area; external motion data is collected by infrared positioning device; in the treatment of in vitro real-time motion data, and through the relevant model to calculate position information of the target area; and then through the target prediction algorithm to predict the movement, finally the results of the model is transferred to the treatment system to adjust treatment. But the Synchrony system also has some shortcomings, such as the invasive implantation procedure, The error of the correlation model and the prediction model is very large.
The tracking model of Synchrony system as the basis, we propose a tracking model of respiratory motion external motion signal based on the model structure and similar Synchrony system, but the realization of each module are different. In the model, the in vitro motion data collected by POLARIS infrared positioning system NDI. Communication interface function this paper is based on NDI offering, to achieve a practical in vitro respiratory motion measurement system, the main functions include motion data acquisition, display and recording, marker position is displayed in the three field of view, the real-time calculation of respiratory motion parameters of respiratory motion is predicted through prediction algorithm, display real-time motion curve. The prediction curve and prediction error. In the experiment, the motion data of infrared reflective markers placed on the patient's body surface as in vitro respiratory motion data.
The body motion data collected by digital simulator, the motion information of diaphragm as body motion data. In the perspective of mode, through the camera to record the motion of the diaphragm, and then converted to digital video image, the target tracking algorithm automatically get position information of the diaphragm in the image. This paper implements three a target tracking algorithm: minimum absolute deviation algorithm (MAD algorithm), maximum close distance algorithm (MCD algorithm) and mutual information algorithm (MI algorithm), the results show that the three algorithms can track the moving target effectively, the MI algorithm the accuracy and robustness of the best, and too slow to solve the problem of matching speed of MI algorithm, using a step search method to search and accelerate the process.
In the treatment process, motion tracking strategy is applied in low frequency sampling, high frequency sampling, from in vitro calculated body movement, the correlation model therefore requires the establishment of in vivo and in vitro in motion at the start of treatment, by tracking the position information by in vitro movement to know the target area in the treatment. But from the motion tracking device tracking to make adjustments between the treatment equipment system delays, including data acquisition time, computing time, data transmission and mechanical time delay, total delay time can reach hundreds of milliseconds, processing method of this problem is the most effective location information through the prediction model to predict the target area.
The correlation model, the input data is the external motion, target motion estimation data as output; and for the prediction model, is the current value as input, the future value as output, the nature of the two models are very similar, so we can use the same function form, and then the corresponding model is given according to different input and output calculate the corresponding model parameters, respectively, to construct two models. But due to respiratory motion itself is very irregular, and the accuracy of radiotherapy for model and real-time requirements are very high, it is difficult to meet the requirements of the use of traditional modeling methods, this study proposes the application of nonparametric correlation model and regression forecasting model construction method.
This collection of respiratory motion data of 11 subjects, and then uses the non parametric regression model, regression model and BP neural network prediction model, and compared with the results without prediction. At the same time for the prediction process of the "abnormal state", proposed a modified nonparametric regression prediction methods. Finally prediction algorithm is integrated in the measurement system, the validity of the forecast algorithm in real time measurement to test. The test results show that, in the prediction of different lengths, can accurately predict the respiratory motion of non parametric regression method, the improved method can greatly reduce the respiratory motion in the abnormal state prediction the error in prediction. The length of 0.6s, 11 sets of data in prediction, regression model, BP neural network, nonparametric regression and the improved regression method the normalized mean square error of the mean was 0.85,0.54, 0.52,0.44 and 0.4, and combined with the measurement system, the algorithm can also be predicted in real time and accurately.
The correlation model of Synchrony system used in the mixed polynomial model, which has the advantages of simple structure, but the model error in the experiment, we established a correlation model based on nonparametric regression method, and compared with the linear model, two biquadratic polynomial model and neural network model. The results were validated using 7 group internal and external synchronous motion data, in which the body motion data for the motion data obtained by 3D ultrasound blood vessels in the liver, the motion data in vitro data for surface markers obtained by optical measurement system. Through calculation, linear, double two degree polynomial, four correlation models and nonparametric regression neural network were normalized the mean square error was 0.35,0.32,0.30,0.19, so the error correlation model based on nonparametric regression is much smaller than the other three models, and the model construction of convenient calculation The real time is good.
In vitro tracking movement through the optical device, motion information can simultaneously track multiple markers, more sampling points, in vitro data contains more, this model will be more complicated. In the study, this paper established the in vitro - nonparametric regression model based on the correlation of body movement, the use of 1. The 2 model, the normalized mean square error mean respectively 0.185,0.136,0.126, 3 in vitro markers when the number of marker in vitro model including the model error is small, but the combination of error and the marker in vitro model of certain relationship does not exist.
Inside and outside the motion correlation model was established by comparison of different algorithms, the motion prediction model and in vitro - in vivo correlation model, it has strong robustness in modeling the respiratory movement in non parametric regression method, high accuracy, the advantages of good real-time performance, can meet the requirements of real-time tracking and tracking with radiotherapy. The. The expansion history database model continuously, the accuracy of the model will be improved.
At the end of the paper, we make a summary of the work done, at the same time, explain some of the shortcomings in the article and make some prospects for the future work.

【学位授予单位】：南方医科大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41;R318.0

【参考文献】