Data reduction for cough studies using distribution of audio frequency content
1 School of Electrical and Electronic Engineering, University of Manchester, Manchester, United Kingdom
2 Respiratory Research Group, Manchester Academic Health Sciences Centre, University of Manchester, University Hospital of South Manchester, ERC Building, Second floor, Manchester, M23 9LT, United Kingdom
Cough 2012, 8:12 doi:10.1186/1745-9974-8-12Published: 12 December 2012
Recent studies suggest that objectively quantifying coughing in audio recordings offers a novel means to understand coughing and assess treatments. Currently, manual cough counting is the most accurate method for quantifying coughing. However, the demand of manually counting cough records is substantial, demonstrating a need to reduce record lengths prior to counting whilst preserving the coughs within them. This study tested the performance of an algorithm developed for this purpose.
20 subjects were recruited (5 healthy smokers and non-smokers, 5 chronic cough, 5 chronic obstructive pulmonary disease and 5 asthma), fitted with an ambulatory recording system and recorded for 24 hours. The recordings produced were divided into 15 min segments and counted. Periods of inactive audio in each segment were removed using the median frequency and power of the audio signal and the resulting files re-counted.
The median resultant segment length was 13.9 s (IQR 56.4 s) and median 24 hr recording length 62.4 min (IQR 100.4). A median of 0.0 coughs/h (IQR 0.0-0.2) were erroneously removed and the variability in the resultant cough counts was comparable to that between manual cough counts. The largest error was seen in asthmatic patients, but still only 1.0% coughs/h were missed.
These data show that a system which measures signal activity using the median audio frequency can substantially reduce record lengths without significantly compromising the coughs contained within them.