Respiratory motion modelling and prediction using probability density estimation

One of the current major challenges in clinical imaging is modeling and prediction of respiratory motion, for example, in nuclear medicine or external-beam radio therapy. This paper presents preliminary work in developing a method for modeling and predicting the temporal behavior of the anterior surface position during respiration. This is achieved by tracking the anterior surface during respiration and projecting the captured motion sequence data into a lower dimensional space using Principle Component Analysis and extracting the variation in the Abdominal surface and Thoracic surface separately. Modeling is based on learning the multivariate probability distribution of the motion sequence using a joint Probability Distribution Function (PDF) between the variation of the Thoracic surface and Abdomen surface in the Eigen space. Moreover, the prediction model encodes the amplitude of the variation in the Eigen space for both Thoracic surface and Abdominal surface and the derivative of the variation which reflects the motion path (velocity). The joint Probability Distribution Function (PDF) of the prediction model covers the likelihood of each position/phase configuration and the associated maximum-likelihood motion path. Moreover, feeding the real-time tracking data into the model during nuclear medicine acquisition or external-beam radio therapy will facilitate adjusting the model for any changes and overcome irregularities in the observed respiration cycle.