Study On Respiratory Motion Tracking Model Based On External Signal

Radiotherapy plays one of the most important roles in oncology; its purpose is to deliver the maximum radiation dose to the target while keeping the dose to the surrounding normal tissues below tolerance. With the development of technology, various kinds of precise radiotherapy techniques have emerged, survival rates and quality of life of the patients have been steadily improved.But there are still many uncertainties in radiotherapy, such as inter-fraction set-up errors, the target deformation during the treatment, as well as various kinds of non-autonomous motions, especially, respiratory motion which will change the location of tumors in the thorax and abdomen significantly. The effect of respiratory motion would arise in the whole process of radiotherapy, including treatment plan images acquisition, plan designing and plan delivery. Conventional methods to manage respiratory motion in radiotherapy including:motion-encompassing, forced shallow-breathing, breath-holding and respiratory gating, however, these methods can’t manage the respiratory motion satisfying. Real-time tumor-tracking method is the best known method which used in respiratory motion management; the tracking devices obtain the targets position in real time, and then feedback this information to the beam adjusting device, reposition the radiation beam to follow the tumor’s position in real-time and accurately.One of the most important problems in real-time tumor-tracking radiotherapy is how to determine the moving targets positions accurately in real-time. The most widely used methods including:detecting artificial fiducials that are implanted in or near the tumor by radiography, tracking surrogate organs that move in synchrony with tumor by radiography and tracking the motion of the makers that fixed on the patients’surface by optical measuring devices. Tracking the target by radiography imaging would be accurately but patients will accept too much imaging dose, and the procedure of implanting fiducials is invasive; the optical tracking methods can gets patients’external motions in real time, the procedure is very convenient and is harmless to patients, but as the inconstant correlation between the internal and the external motion, it is difficult to track the internal target accurately from external signal. Currently, the most feasible method is to use internal and external tracking methods simultaneously, and take the advantages of both to detect targets’motion. The synchrony respiratory tracking system (Synchrony system, Accuray Inc) which has been widely used in clinic is based on this idea. With two digital X-ray generators and amorphous silicon X-ray detectors, Synchrony system tracking the implanted fiducials to obtain the internal motion information; and the external motion data is collected by infrared positioning system; during treatment, internal target position is estimated from external makers positions which is obtained in real time by internal/external correlation model; at last, the predicted position is sent treatment system to adjust the treatment system. But the Synchrony system also exist some deficiencies, such as invasive implantation procedure and the accuracy of correlation model and the prediction model is not satisfying.We proposed an improved respiratory motion tracking model base on external motion signal, the model structure is similar to Synchrony system, but the modules are different. In the model, the external motion data are collected by POLARIS (Northern Digital Inc) infrared positioning system. On basis of the application programming interface provided by NDI, we completed a practical external respiration motion measurement system, the main function including:motion data acquisition, displaying and recording, displaying the makes in three orthographic views of view field, calculating respiratory parameters and predict the respiratory motion in real time, and showing motion curves, prediction curves and prediction error curves in real time. In the experiments, the external respiratory motion data is the motion signal of the infrared passive makers place on the patients’ surface.The internal motion is the motion signal of diaphragm which is acquired by digital simulator. In the fluoroscopy mode, motion of the diaphragm is recorded by camera, and then be turned into digital images, the top of diaphragm is detected by target tracking algorithm in the images automatically. In experiments, three target tracking methods have been completed, including:minimum absolute deviation algorithm (MAD algorithm), maximum close distance algorithm (MCD algorithm) and mutual information algorithm (MI algorithm), results indicates that all the three algorithm can track moving target effectively, and the most accurate and robust one is MI algorithm, at last, an equal-step search method is proposed to speed up the search of MI algorithm.In radiotherapy, respiratory motion tracking strategy is low external sampling rate and high internal sampling rate, estimating the internal motion from external motion; therefore, it is necessary to build the correlation model before treatment. And there is always a system latency due to the time need in data acquisition, data processing, communication delays, mechanical latency, and so on, the total latencies would be hundreds of milliseconds or more, The most convenient method to compensate the system latencies is to predict the target position by in advance.In correlation model, the input data is the external motion signal and the output data is the estimated target position; in prediction model, the input data is the current data, and the output is future data, as the structure of the two models are very similar, so the same form of function can be used, given corresponding inputs and outputs, calculate the parameters of models, at last, different models are built. However, due to the irregularities of respiration and the requirement of accuracy and real-time in radiotherapy, the results obtained from traditional methods were hardly acceptable. We proposed to use the nonparametric regression method in building the respiratory motion correlation model and prediction model.The respiratory trajectory of11volunteers were collected and predicted based on nonparametric regression method. The results were compared with those of autoregressive and neural network. An improved method was proposed to deal with the abnormal state in respiration. Furthermore, we combined the prediction method with the tracking system to test its performance in practical application. The results indicated that the proposed method could predict the motion accurately in real-time for different latencies, the mean of normalized root mean square error (nRMSE) was0.85,0.54,0.52,0.44and0.4for no prediction, autoregressive, neural network, nonparametric regression and the improved method with a prediction length of0.6s. This method also could predict respiration signals effectively when combined with the positioning system.The correlation model used in the Synchrony system is blending polynomial model, its structure is simple, but the accuracy is not satisfying, in experiments, we built correlation model based on nonparametric regression method, and the results were compared with linear model, dual quadratic polynomial model and neural network model. Seven internal-external synchronized data sets were used to evaluated those models, the motion of vessel bifurcations in the liver were recorded as target motion by3D ultrasound, the makers fixed on the volunteers’surface were recorded as external motion by optical positioning system. The mean of normalized root mean square error of linear, dual quadratic polynomials, neural networks and nonparametric regression correlation model were0.35,0.32,0.30and0.19respectively, the nonparametric regression method with minimum error which is much lower than other three, furthermore, its structure is very simple and could estimate the results in real time.When tracking the external motion by optical device, several makers could be tracked simultaneously, the more makers be tracked, the more data would be collected, and the correlation model contains multiple makers would be more complex. In our research, we built multi-external/internal correlation model based on nonparametric regression method, the normalized root mean square error of models contained one, two, and three external makers were0.186,0.136and0.126respectively. The result indicated that more makers the model contains, lower errors would be, but the relationship between the accuracy and the combination of external makers was uncertain.Compared the results of internal/external correlation models, prediction models, and multi-external/internal correlation model base on different algorithm, we know that the nonparametric regression method used in respiratory motion models is very robust, accurate, and real-time and which could meets the requirements of real time tumor tracking radiotherapy. Furthermore, the model would be more accurate with the updating of historical database.At the end of this thesis, we summarize the work briefly, and discuss some problems which haven’t been completed. In addition, we prospect some work to do in the future.