A LIGHTWEIGHT MOBILENETV2-BASED FRAMEWORK FOR PNEUMONIA DETECTION FROM CHEST X-RAY IMAGES

Pneumonia is a major cause of mortality worldwide, particularly in regions with limited access to expert radiologists. Deep learning has shown strong potential for automated analysis of chest X-rays, but many CNN models are too heavy for deployment in real-time or resource-constrained settings. This paper presents a lightweight deep learning framework based on transfer learning with MobileNetV2 for binary classification of chest X-ray images into Normal and Pneumonia classes. The pipeline includes systematic dataset splitting, task-specific data augmentation, a frozen MobileNetV2 backbone with a compact classification head, and evaluation using accuracy, precision, recall, confusion matrix, and ROC–AUC. Experiments on a dataset of 5856 chest X-ray images (4555 train / 600 validation / 701 test) yield a test accuracy of 89.02%, precision 92.53%, recall 91.97%, and ROC–AUC 0.95. Qualitative inspection of best-classified and misclassified images shows that the model is robust on typical cases but challenged by ambiguous or low-quality images. The proposed solution is suitable for integration into clinical decision-support tools and edge devices.

Hali tarjima qilinmagan