Pneumonia detection and classification from chest X-Ray images using Modified DenseNet121 Deep Learning Approach

Pneumonia is a potential life-threatening lung infection that remains a global health concern even today. Adequate diagnosis in time is essential to improve therapeutic results. In this study, we present a modified deep learning-based approach "transfer learning" for automatic identification of pneumonia in chest X-ray images. Utilizing transfer learning with pre-trained DenseNet121 model and leveraging the feature extraction capabilities, we fine-tuned them with additional connected layers specifically tailored for binary classification to distinguish between pneumonia normal and positive cases. We have applied data augmentation techniques to expand the small dataset and enhance model generalization. The modified model is trained with training and validated with respective datasets. Our proposed modified DenseNet121 model achieved classification accuracy of 90.06%, with a sensitivity of 0.89 and specificity of 0.90 with testing datasets. These metrics indicate that the model is highly effective in identifying pneumonia cases from X-ray images. The proposed approach has remarkable potential for clinical use, providing a reliable, automated solution for assisting radiologists in diagnosing pneumonia, which could help reduce the diagnostic burden in resource-constrained environments and improve patient outcomes. Overall, this method combines the power of transfer learning, augmentation, and fine-tuning to deliver higher accuracy and offers a way for quick preliminary automated pneumonia detection.