Internal-state analysis in a layered artificial neural network trained to categorize lung sounds

In regular use of artificial neural networks, only input and output states of the network are known to the user. Weight and bias values can be extracted but are difficult to interpret. We analyzed internal states of networks trained to map asthmatic lung sound spectra onto lung function parameters. Decorrelation of the spectral data revealed that the spectra can be seen as composed of distinct intracorrelated frequency bands. The effective pitch shifts with increasing degree of airways obstruction. By comparing internal state analysis and decorrelation analysis, we concluded that our neural network performs a simulation of a decorrelation operation.