A Proof-of-Concept Study for the Real-Time Prediction of Respiratory Patterns: a Simple Bayesian Approach

Abstract

Recent radiation therapy has overcome the effect of internal organ motion by limiting the range of movements, gating the beam irradiation or tracking the target movement. A successful strategy requires accurate real-time estimation of target location during radiation treatment. In this study, we propose a relatively simple technique to predict patient’s respiratory pattern (RP) one step before the breathing using a Bayesian approach. Patients’ respiratory signals (RSs) were analyzed using the in-house RPM signal analyzer, and parameters (period (τ), baseline (β) and amplitude (ϕ)) characterizing an RP were extracted. Based on each parameter, we obtained the probability density function (PDF) and transition probability matrix defined as ‘likelihood’. We predicted the following RP based on the PDF and the likelihood, then compared the estimated RP with the actual one. The proposed method was applied to five lung cancer patients who were treated with radiation therapy in our facility. Prediction error was analyzed using root-mean-square error (RMSE;e, in mm) and relative RMSE (ϵ, in %) for each breathing cycle in all RP. The e range was [0.45,2.66], and ϵ range was [5,18.8]. The prediction accuracy was strongly dependent on the irregularity of RP. Although the prediction errors were more significant than expected, we could confirm the feasibility of the proposed algorithm. The proposed algorithm is more intuitive than other sophisticated methods and requires less computation time.