Predictive feedback control of the treatment couch for tumor motion compensation during radiotherapy

In radiation therapy, tumor motion induced by the patient's respiration may lead to significant differences between the planned and delivered radiation dose. Compensating for tumor motion is therefore crucial for accurate and efficient treatment. In this dissertation, we study a real-time compensation method through predictive feedback control of the treatment couch.;Conventional approaches (margin expansion, breath-holds, gating) have substantial drawbacks. Real-time tracking has the potential to overcome these limitations. A real challenge in real-time tumor motion tracking approach is the presence of delays in the treatment system. In this dissertation we studied a new real-time tracking method, which compensate the tumor motion through feedback control of the treatment couch. This approach consists of first applying a predictor to overcome the delays and then the predicted signal is used as the reference for the controlled couch. In order to handle the irregularities of the breathing signals, we developed and evaluated optimized versions of the Least Mean Squares (LMS), the Recursive Least Squares (RLS) algorithm, and Neural Networks (NN). Additionally, the use of a Nonlinear Set Membership (NSM) algorithm for prediction of human breathing was investigated. This approach does not require the choice of a predefined functional form for the predictor. The NSM approach resulted in better prediction performance compared to optimized LMS, RLS, and NN. In addition the convergence issue appeared to be completely solved. Considering the predicted signal to be the tracking reference, we then pursued the problem of predictive feedback real-time tumor motion tracking, by designing the couch controller following two different approaches, a pole placement method and an optimal control control method. The pole placement method was found to be insufficiently robust due to inadequate stability margin. The optimal approach appeared to be more robust and gave optimal tracking results. The real-time tracking approach was evaluated by performing simulation using nine real clinical signals. Only two of the nine cases appeared to have tracking error samples exceeding 0.3 cm and at most 1.14% of the tracking error samples exceeded the 0.3 cm threshold. These results show the validity of the proposed approach as a real-time tumor motion tracking solution.