Objective: In this work, we introduce a quantitative non-contact respiratory evaluation method for fine-grain exhale flow and volume estimation through Thermal- CO2 imaging. This provides a form of respiratory analysis that is driven by visual analytics of exhale behaviors, creating quantitative metrics for exhale flow and volume modeled as open-air turbulent flows. This approach introduces a novel form of effort-independent pulmonary evaluation enabling behavioral analysis of natural exhale behaviors.
Methods: CO2 filtered infrared visualizations of exhale behaviors are used to obtain breathing rate, volumetric flow estimations (L/s) and per-exhale volume (L) estimations. We conduct experiments validating visual flow analysis to formulate two behavioral Long-Short-Term-Memory (LSTM) estimation models generated from visualized exhale flows targeting per-subject and cross-subject training datasets.
Results: Experimental model data generated for training on our per-individual recurrent estimation model provide an overall flow correlation estimate correlation of R2=0.912 and volume in-the-wild accuracy of 75.65-94.44%. Our cross-patient model extends generality to unseen exhale behaviors, obtaining an overall correlation of R2=0.804 and in-the-wild volume accuracy of 62.32-94.22%.
Conclusion: This method provides non-contact flow and volume estimation through filtered CO2 imaging, enabling effort-independent analysis of natural breathing behaviors.
Significance: Effort-independent evaluation of exhale flow and volume broadens capabilities in pulmonological assessment and long-term non-contact respiratory analysis.