Enhanced Deep Learning Framework for Lung Disease Detection Using Optimized Segmentation and Classification of Chest X-Ray Images

The automated detection of pulmonary diseases by Chest X-ray (CXR) images is crucial in modern medical diagnostics, enabling early identification and improved patient outcomes. Precise segmentation of pulmonary areas is crucial for improving classification accuracy, reducing false positives, and increasing interpretability in deep learning models. This research introduces an improved U-Net segmentation technique, U-Net-R50-ResB3-LReLU, which integrates a ResNet-50 encoder featuring three residual blocks and Leaky ReLU activation to enhance feature extraction and segmentation accuracy. The effectiveness of various U-Net architectures is evaluated using the Dice Coefficient, Jaccard Index, Sensitivity, and Specificity, demonstrating that the proposed model exceeds the performance of the baseline U-Net and its variants. The pulmonary regions are classified into four categories: Normal, COVID-19, Lung Opacity, and Viral Pneumonia, employing deep learning architectures including ResNet50, DenseNet201, VGG19, InceptionResNetV2, and EfficientNetB7. The experimental results indicate that the U-Net-R50-ResB3-LReLU segmentation model, combined with EfficientNetB7, achieves the highest specificity (97.50%) and F1-score (96.02%), exceeding conventional segmentation techniques. These findings emphasise the significance of accurate segmentation in improving the reliability of automated CXR analysis, hence enabling more precise and efficient detection of pulmonary illnesses in clinical settings.