Pneumograph Signal Processing and Feature Extraction

Respiration line length, measured as excursion, was studied using N = 72 segments of data from 36 stimulus presentations sampled from comparison questions tests conducted during confirmed field investigations. Raw data were compared to signal processing models including low-pass filtering and interpolation of answer-movement artifacts. Measurements using raw data explained 2.4% of the criterion variance, while processed data explained 15.6% of the variance; a difference that was statistically significant (p = .018). After combining the data from thoracic and abdominal sensors, excursion measurements using the processed respiration data concurred with the binary case status for 69.4% of the sample segments, with a criterion coefficient that accounted for 18.5% of the variance in case status. Results using raw data concurred with the criterion state for 61.1% of the sample segments, and produced a criterion coefficient that accounted for 4.8% of the variance in case status. Scores using the filtered data concurred with the criterion status at a rate that was significantly greater than chance (p = .007), while results using the raw data were not significantly different than chance (p = .086). Sample data indicate that high sampling rates can introduce non-diagnostic noise that significantly reduce the diagnostic value of respiratory excursion measurements, while processing of raw pneumograph data can improve the accessibility of diagnostic information.