Real-Time Markerless Respiratory Motion Management Using Thermal Sensor Data

摘要

We present a novel approach to jointly generate multidimensional breathing signals and identify unintentional patient motion based on thermal torso imaging. The system can operate at least 30% faster than the currently fastest optical surface imaging systems. It provides easily obtainable point-to-point correspondences on the patients surface which makes the current use of computationally heavy non-rigid surface registration algorithms obsolete in our setup, as we can show that 2d tracking is sufficient to solve our problem. In a volunteer study consisting of 5 patient subjects we show that we can use the information to automatically separate unintentional movement, due to pain or coughing, from breathing motion to signal the user that it is necessary to re-register the patient. The method is validated on ground-truth annotated thermal videos and a clinical IR respiratory motion tracking system.