Template-based respiratory monitoring and tidal volume estimation.

Abstract

Vision-based monitoring has emerged as a promising alternative to traditional contact-based solutions in intensive care units. Though existing studies have shown its potential in respiratory monitoring, this approach is still in its early stages. For instance, the impact of the body region to be monitored has rarely been investigated, especially in respiratory volume estimation. Furthermore, important clinical constraints including body motion and occlusion artifacts are usually neglected for simplification. These drawbacks compromise the practicality of vision based respiratory monitoring in clinical uses. In this re-search, we propose a novel approach to modeling the torso region, which is instrumental in accurate respirator monitoring and volume estimation. First, the patient's body surface is captured using a Kinect camera. The body surface is then aligned and registered with a pre-defined human template, allowing consistent modeling and segmentation of the respiration-relevant region. Finally, an efficient yet accurate volume computation algorithm is initiated for volume monitoring and respiratory parameter extraction. The effectiveness of our proposed approach is validated on a group of voluntary subjects by comparing it with gold standard spirometry as well as with annotation by experts. Overall, the system reaches an accuracy of over 90% for tidal volume estimation and 95% for respiratory rate estimation without requiring patient-specific calibration data, thus illustrating its effectiveness and robustness for ICU respiratory monitoring.