Resource-efficient diverse lung disease classification using lite-convolutional neural networks

A precise and effective classification of lung diseases is of crucial importance for early diagnosis and effective treatment. In this study, we propose a lite neural network (Lite-CNN) for the classification of lung diseases with various classes that use Mel spectrograms from audio signals with a respiratory tract. Unlike conventional image methods, our approach transforms lung recordings into performance of the Mel spectrogram, so that the model can record complex frequency characteristics associated with different respiratory conditions. The Lite-CNN proposal is optimized for the effectiveness of the calculation and at the same time maintains the precision of the high classification, which means that it is suitable for medical applications in real time. The model was trained and evaluated in a different set of data with respiratory audio signals that correspond to a chronically obstructive pulmonary disease (MPOC), pneumonia, heart failure, effusion of pleura, asthma and lung fibrosis. Thanks to the depth experiences, which include the advanced extraction of functionality, the expansion of data and the regulation of the hyperparameter, the model has obtained an exceptional precision of the classification of 99%. This high performance shows the effectiveness of Lite-CNN in the distinction between different lung diseases and, at the same time, requires much less arithmetic resources than conventional depth learning models. The integration of Mel spectrograms improves characteristic extraction, which leads to better diagnostic precision. This study highlights the potential of controlled diagnostic tools with light artificial intelligence in the classification of respiratory diseases and paid the path of low and accessible health solutions. Future research focuses on real implementations and clinical validation to further improve practical applicability.