Computer-Aided Lung Auscultation Screening and Radiographic Evaluation of Pediatric Pneumonia

While chest auscultations provide an accessible and low-cost tool for pediatric pneumonia diagnosis, its subjectivity and low reliability continues to hinder its inclusion in global pneumonia guidelines; eventhough more robust tools like chest radiography also suffer from cost and accessibility issues. Advances in computer-aided analytics is offering more robust tools for interpreting digital auscultation signals though little has been done to explore variations of lung sounds across different chest positions and the correspondence between auscultations and specific radiographic findings. The present study explores interpretation of lung auscultations across chest positions in a pediatric pneumonia population, using a deep neural network classification of normal and abnormal breathing patterns. The results reveal a strong alignment between computer-aided auscultation findings and radiographic interpretation not only in terms of presence of adventitious lung patterns, but also in terms of localizing the abnormality along the left or right lung. Though evaluated in a small clinical population, this research underscores the potential of computer-aided auscultation analysis as a cost-effective substitute for radiography in resource-limited settings.