Applications of the Analysis of Respiratory Kinematics (ARK) System to Respiratory Distress Outcome Prediction

Clinical respiratory monitoring has been historically limited by the available technology that can reliably produce useful metrics without jeopardizing patient comfort. Pulmonary condition can be accurately assessed by physicians who have the training to identify relevant warning signals from visual examinations and available metrics such as average respiratory rate and oxygen saturation. Traditional methods of laboratory pulmonary function testing are powerful and diverse, but lab procedures would generally be infeasible for widespread application due to the cumbersome setup and equipment and the time required to take measurements. To combat this translational research gap, new methods using lightweight technologies like wearable sensors have proliferated, demonstrating large potential for quantifying respiratory status. While many methods have been proposed, few have demonstrated the ability to monitor for multiple signs of respiratory distress, and similarly few have demonstrated sufficient robustness for clinical use. The latter issue holds especially true for inertial sensors tracking the motions of breathing, as they also detect gross body motions as noise that must be filtered. Still, the compact designs of modern inertial sensors are an ideal form factor for both clinical and ambulatory care settings.

To address this set of research questions, we designed the Analysis of Respiratory Kinematics (ARK) system to quantify clinically-relevant metrics of respiratory health with the goal of predicting negative patient outcomes. ARK employs multiple motion sensors with inertial measurement cores at key anatomical locations on the chest and torso, tracking motion along the chest wall. ARK applies novel methods for analysis of respiratory data based on robust signal processing techniques previously verified in other clinical arenas, extracting known markers like a noise-filtered respiratory rate time series and resulting rate variability, as well as novel quantitative metrics for warning signs like the recruitment of accessory muscles and respiratory alternans. To facilitate design of clinically useful metrics, ARK has been deployed in a medical exercise lab, emergency department, and hospital wards and intensive care units. Preliminary predictive analysis shows that ARK produces metrics that can discriminate between patient outcomes and have statistical value for outcome modeling. This work represents one of the first applications of robust, multiparametric respiratory information derived from inertial sensors to generalized clinical outcome modeling and prediction, validating the potential for inertial respiratory sensing even in noisy conditions.