Assessment of input signal positioning for cardiac respiratory motion models during different breathing patterns

Abstract

Motion models have been applied as a solution to the problem of respiratory motion in a range of applications. Such models predict motion fields based on 1-D signals or signal combinations. These signals often measure the motion of a region of the subject's anatomy, such as the chest surface or diaphragm. The hypotheses we investigate in this paper are that the predictive accuracy of motion models will vary depending on the choice of input signal(s) used by the model, and furthermore that the optimal choice of signal(s) will vary depending on the breathing pattern of the subject (e.g. normal breathing, deep breathing, fast breathing). We test these hypotheses by forming cardiac respiratory motion models from dynamic MRI data acquired from 9 volunteers. For input signals we produce post-processed 'virtual navigators' from the dynamic MRI images, enabling us to test arbitrary navigator positions and orientations. Our results support both of our hypotheses. We show that the optimal choice of input signal over all breathing patterns was a combination of signals including one positioned on the diaphragm and either one on the abdominal surface or one on the lateral wall of the heart. In addition, the best combination changed as the subject altered their breathing pattern.