Diagnosis of Community-Acquired pneumonia in children using photoplethysmography and Machine learning-based classifier

This paper presents a novel approach for diagnosing Community-Acquired Pneumonia (CAP) in children using single-channel photoplethysmography (PPG) using machine learning Traditional diagnostic methods (x-rays systems and blood tests) for pneumonia suffer from limitations e.g., unavailability in remote rural areas, time consumption, financial burden, and reliance on invasive procedures. This novel approach uses the PPG recording alone to generate accurate and rapid diagnoses of CAP in children that may facilitate healthcare practitioners in low-resource clinical settings in future. A cross-sectional study was carried out to collect the PPG recordings of 67 paediatric participants (31 CAP and 36 healthy). Five different machine learning classifiers namely Fine Decision tree, Linear Discriminant Analysis, Weighted K Nearest Neighbour, Wide Neural Network, and Ensemble of Bagged Trees using eight PPG signal features were employed. Using weighted KNN we predicted 9 out of 10 test subjects correctly. These results demonstrate the potential of the system to improve clinical decision-making and patient outcomes since despite the thriving advancements in healthcare paediatric pneumonia remains a major health concern.