Classification of asthmatic breath sounds: Preliminary results of the classifying capacity of human examiners versus artificial neural networks

For continuous monitoring of the respiratory condition of patients, e.g., a t the intensive care unit, computer assistance is required. Existing mechan ical devices, such as the peak expiratory flow meter, provide only with inc idental measurements. Moreover such methods require cooperation of the pati ent, which at, e.g., the ICU is usually not possible. The evaluation of com plicated phenomena such as asthmatic respiratory sounds may be accomplished by use of artificial neural networks. To investigate the merit of artifici al neural networks, the capacities of neural networks and human examiners t o classify breath sounds were compared in this study. Breath sounds were in vivo recorded from 50 school-age children with asthma and from 10 controls . Sound intervals with a duration of 20 seconds were randomly sampled from, asthmatics during exacerbation, asthmatics in remission, and controls. The samples were digitized and related to peak expiratory flow. From each inter val, two full breath cycles were selected. Of each selected breath cycle, a Fourier power spectrum was calculated. The so-obtained set of spectral vec tors was classified by means of artificial neural networks. Humans evaluate d graphic displays of the spectra. Human examiners could not clearly discri minate between the three groups by inspecting the spectrograms. Classificat ion by self-classifying neural networks confirmed the existence of at least three classes; however, discrimination of 11 classes seemed more appropria te. Good results were obtained with supervised networks: as much as 95% of the training vectors could be classified correctly, and 43% of the test vec tors. The three patient groups, as discriminated in advance, do not corresp ond with three sharply separated sets of spectrograms. More than three clas ses seem to be present. Humans cannot take up the spectral complexity and s howed negative classification results. Artificial neural networks, however, are able to handle classification tasks and show positive results, (C) 199 1 Academic Press.