Validation of a Digital Stethoscope for Recording and Identifying Pediatric Breath Sounds

Introduction: Auscultation of breath sounds at the bedside remains an essential part of physical exams, especially in pediatric patients as respiratory illness accounts for 17-20% of inpatient pediatric hospital admissions annually. Identification of specific breath sounds, including wheezing, crackles and rhonchi, may guide the use of antibiotics and bronchodilator therapies. Using a digital stethoscope (Eko) for recording, we created a database of pediatric breath sounds, both normal and abnormal, that may be utilized for generation of a machine learning algorithm.


Methods: This is a prospective observational study conducted at a tertiary pediatric center, including patients aged 0 through 18 years old admitted to the general inpatient pediatrics wards and the pediatric intensive care unit (PICU). Excluded patients were those admitted to the cardiac intensive care unit or the neonatal intensive care unit. Enrolled patients had breath sounds auscultated and recorded in 10 locations (5 anterior and 5 posterior). Breath sound classification was assigned by the recorder at the bedside and by 5 expert listeners who evaluated the recordings.


Results: Cohen’s kappa values were calculated to determine degree of agreement in assigning breath sound classifications to recorded sounds as compared to the classifications by the bedside recorder. Interrater agreement is moderate (k = 0.43) when comparing bedside recorder classification to majority classification of all listeners. All five listeners when compared individually to the recorder demonstrated fair agreement. There was no difference in agreement based on age, gender, respiratory support, nebulizer use (fair) or specific breath sound (as assigned by the bedside recorder).


Conclusions: Eko digital stethoscope recording technology can be used to record pediatric breath sounds with fair to moderate agreement in identifying sounds when compared to bedside auscultation. This database may allow for the generation of a machine learning algorithm to automatically identify pediatric breath sounds at the point of care. Recordings may also be a useful teaching tool for pediatric respiratory findings and disease processes.