Classification of lung sounds in patients with asthma, emphysema, fibrosing alveolitis and healthy lungs by using self-organizing maps.

The performance of the self-organizing map (SOM), an artificial neural network, was evaluated in the classification of lung sounds. Patients with asthma (n = 8), emphysema (n = 8) and fibrosing alveolitis (n = 8), and patients with healthy lungs (n = 8) were selected for the study. Fast Fourier transform (FFT) spectra from midinspiratory breath sounds recorded at the right lower lobe area were used to construct feature vectors in the learning and classification process of SOM. The sound segments did not contain wheezing sounds. The lung sounds of 25/32 (78%) patients were classified correctly, with an overall kappa (kappa) value of 0.71. The agreement between the clinical and proposed diagnoses based on classification of lung sounds was good among patients with emphysema (kappa = 0.92) and those with healthy lungs (kappa = 0.83), but only moderate among patients with asthma (kappa = 0.52) and fibrosing alveolitis (kappa = 0.54). This is due to the limitations in distinguishing breath sounds of asthmatics without wheezing sounds from those with crackles in fibrosing alveolitis by the spectral pattern alone. The results indicate that SOM based on FFT spectra is potentially useful in the classification of lung sounds, e.g. in health screening or in differential diagnosis of pulmonary disorders. To enhance the performance of SOM, other features of lung sounds should be combined with FFT spectra.