PEDIATRIC CHEST DISEASE PREDICTION

Abstract

This study addresses the classification of pediatric chest X-ray images into three diagnostic categories: Infiltration, Effusion, and Atelectasis. Utilizing the NIH Chest X-ray dataset filtered for patients aged 12 or younger, we employ convolutional neural networks (CNNs) for feature extraction and rigorously compare various machine learning and deep learning models, including Random Forest, Logistic Regression, SVM, Decision Tree, and DenseNetl21. A key contribution of our research is the use of a pre-trained VGG16 model for feature extraction, followed by the application of a Random Forest classifier, which achieved a high classification accuracy. Our experimental methodology involves addressing class imbalance through dataset balancing techniques and evaluating models on differently sized datasets. Results indicate that larger datasets substantially enhance model performance, with DenseNetl21 attaining the highest overall accuracy and AUC scores. Notably, our VGG16-augmented Random Forest model also demonstrated robust performance, highlighting the effectiveness of our feature extraction and classification pipeline. The findings of this study underscore the necessity of adequate training data and validate the superiority of deep learning models in handling complex medical imaging tasks. This research underscores the potential of advanced machine learning techniques in improving pediatric chest X-ray classification, thus enhancing diagnostic accuracy and supporting clinical decision-making. Future research directions will explore data augmentation, hyperparameter optimization, ensemble methods, and strategies to improve model interpretability and generalizability.