The Development of a Classifier Based on Neural Networks and K-Neighbors for Pediatric Pneumonia Diagnosis through X-Ray Images

This research focuses on the classification of pediatric pneumonia diagnosis through X-ray images. The database utilized in this study consists of anteroposterior chest X-ray images obtained from retrospective cohorts of pediatric patients aged one to five years at the Guangzhou Women and Children's Medical Center. These images were selected based on their relevance to the study of pneumonia, specifically concerning the identification of bacterial infections.

Using a MATLAB program, seven relevant characteristics were extracted from each X-ray image. These features were essential in determining whether the patient exhibited signs of a bacterial infection associated with pneumonia or if the diagnosis was normal. The classification process was carried out using two distinct methodologies: neural networks and the k-nearest neighbors (K-NN) algorithm. A comparison of these classifiers was performed to evaluate their effectiveness in diagnosing pediatric pneumonia.

The dataset included a total of 49 images diagnosed as normal and 48 images indicating the presence of the bacteria linked to pneumonia. The characteristics considered for analysis included mean, standard deviation, entropy, contrast, correlation, energy, and homogeneity, which play a critical role in image analysis. The results demonstrated an impressive efficiency of 89% for the k-nearest neighbors algorithm and over 96.9% for the neural-network-based classifier, indicating the potential for these methodologies to aid in accurate pediatric pneumonia diagnosis through X-ray imaging.