Abstract:
Early and accurate lung disease prediction and detection is very important for our patient. chestX-ray diagnosis methods are timeconsuming and sesitive process This research exploresdeep learning techniques to automate lung disease classification from chest X-ray images, enhancing diagnostic efficiency and accuracy. We focus on five conditions: Edema, Pneumonia, Tuberculosis, COVID-19, and the normal state.Due to privacy concerns inobtaining X-ray images directly from medical facilities, we utilized a Kaggle, nis.govandv7labs.com datasets of 14,631 chest X-ray images, verified by medical professionals. Thedataset was separated into 80% learning., 10% for testing,10% for assurance. To ensurearobust model, data augmentation techniques such as gamma correction, Image resizing, dataaugmentation, histogram equalization, noise reduction were applied, enhancing the dataset andimproving model performance.We evaluated several CNN model, including (CNN), ResNet50, VGG16, and DenseNet. Each model was assessed based on its training and validationaccuracies. DenseNet is became a very good model, gaining a training accuracy of 99.01%, testing accuracy of 89%, and validation accuracy of 88%, outperforming the other models. VGG16 and CNN also demonstrated high performance, with accuracies around 87%, whileResNet50 gained an accuracy of 80%.DIU Project ReportOur work underscores the potential of advanceddeeplearning models in classifying and identifying lung conditions based on chestX-ray pictures, highlighting a significant improvements in diagnostic efficiency and accuracy that thesetechnologies can offer.
