Application of deep learning methods in the classification of normal and pneumonia lung images.

Background: Pneumonia remains a leading cause of morbidity and mortality worldwide, particularly in vulnerable populations. Rapid and accurate diagnosis through chest radiographs is essential, but manual interpretation can be subjective and time-consuming. This study aims to develop and evaluate deep learning-based models for the automated classification of chest X-ray images into normal and pneumonia categories, providing a foundation for a clinical decision support system. Methods: A retrospective dataset of 500 posterior-anterior chest X-rays (PA) from 2024 was used. Images were preprocessed and resized to 224 × 224 pixels. Four pre-trained convolutional neural network (CNN) architectures-VGG16, ResNet50, InceptionV3, and Xception-were fine-tuned using transfer learning. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the ROC curve (AUC). Results: Among the tested models, VGG16 achieved the highest test accuracy (87.1%) and AUC (0.92), demonstrating strong generalization and classification performance. InceptionV3 also performed well with 85.0% accuracy and AUC of 0.84. The Xception model reached an accuracy of 81.8% (AUC: 0.82) but showed low sensitivity in detecting pneumonia cases. ResNet50 underperformed with an accuracy of 74.2% and AUC of 0.81, likely due to class imbalance and overfitting. Conclusion: VGG16 and InceptionV3 demonstrated high potential for supporting pneumonia diagnosis in chest X-rays. Future research with larger, balanced, and multi-center datasets is needed to improve sensitivity and enhance clinical applicability.