Research Of Respiratory Motion Prediction Based On Wavelet Neural Network

Radiation therapy is an important treatment in treating the liver cancer, and killing cancer cells thoroughly makes it be used widely. In current clinical treatment, chest or abdominal breathing, heartbeat and the movement of diaphragm, severely affected the dose control during treatment. The majority of liver cancer cells require a high dose of radiation to achieve a radical cure, but too high doses of radiation lead to damage to nearby healthy tissues and vital organs. This research arms to get rules of liver tumor by studying its respiration movement, then we propose a prediction approach of respiration motion to fit the long delays during operation.Accurate prediction of liver tumor movement depends on the full understanding of tumor characteristics and the prediction of the tumor movement during treatment. In this study, we analysis large amount of free breathing data, and propose a prediction approach based on Wavelet and neural network. This approach can make an accurate prediction in the long delays, in this paper, we have carried out the fruitful studies in the following aspects:(1) A large number of respiratory motion data for statistical analysis:The basis of my analysis is a finite state-model previously described in Wu et al (2004). A single regular breathing cycle has exactly three breathing states:exhale (EX), end-of-exhale(EOE) and inhale(IN).We defined some notations, like breathing cycle, duration and travel distance, computed the standard deviation and correlation coefficient, and analyzed the inter-patient tumor motion characterization and intra-patient tumor motion characterization.(2) Vitro data collected of the respiratory motion:the data of abdominal will be collected on7volunteers, including three directions, x, y, z. Study how NDI Polaris works and used it in this paper. While collecting data, NDI sends the data to the prediction program in real time. That’s the input of prediction program.(3) We proposed a predict algorithm that fit a longer delays system:study the theory of Haar wavelet and Elman neural network, designed a kind of loosely-structured Wavelet network (WEN) approach, which can meet the requirements in forecasting movement in real time.(4) Experiments and discuss the collected data and the proposed method in this study:compared the WEN algorithm and other methods under the long delay, and root mean square error of WEN can be as low as0.51mm in some case, it is a feasible approach in the clinical.