A Machine Learning Approach on Chest X-Rays for Pediatric Pneumonia Detection

Abstract

Pneumonia is a highly infectious respiratory disease that can be fatal if left untreated. According to the World Health Organization (WHO), pneumonia is the leading infectious cause of death in children younger than 5 years old. Hence, the early detection of pediatric pneumonia is crucial to reduce its morbidity and mortality rate. Even though chest radiography is the most commonly employed modality for pneumonia detection, recent studies highlight the existence of poor interobserver agreement in the chest x-ray interpretation of healthcare practitioners when it comes to diagnosing pediatric pneumonia. Hence, there is a significant need for automating the detection process to minimize the potential human error. Since Artificial Intelligence (AI) tools such as Deep Learning (DL) and Machine Learning (ML) have the potential to automate disease detection, many researchers explored how such tools can be implemented to detect pneumonia in chest x-rays. Notably, the majority of efforts tackled this problem from a DL point of view. However, DL models can be impractical as they possess low medical interpretability. In contrast, ML has been shown to possess a higher potential for medical interpretability while being less computationally demanding than DL. Thus, the objective of this research is to investigate the interpretability of several ML models trained using features extracted from either full or cropped x-rays in order to aid medical practitioners in accurately and reliably diagnosing pediatric pneumonia in chest x-ray images. The performance of these models is compared to a Transfer Learning (TL) benchmark to assess their candidacy. Notably, the results demonstrate that the Logistic Regression (LR) model performs best on cropped images in terms of interpretability while yielding a recall value of 94.07%, which is around 4% less than that of the TL benchmark. However, the added interpretability of the LR model compensates for the slight decrease in model performance when compared to the TL benchmark.