Detection of Lung Disease using Deep Learning Approaches

This electronic document is a “live” template and already Lung ailments are very extensive and are present globally. Common examples of pulmonary illness with chronic obstruction are Asthma, Atelectasis, Consolidation, Cardiomegaly, Edema, Emphysema, Effusion, Fibrosis, Hernia, Pleural, Thickening, fibrosis, Infiltration, Thickening, Nodule, Mass, pneumonia, Pneumothorax, TB and many other diseases. In order to determine these diseases, various machine learning models along with image processing techniques have been established. The proposed model makes use of chest X ray scan techniques to diagnose the infected lungs. Medical professionals can make use of these imaging techniques to study the lungs and decide on the subsequent treatment that has to be taken to cure the ailments at the earliest.In this study,deep learning models are incorporated to offer an excellent method for expert diagnosis of lung illnesses. It aimed to develop a method to aid radiologists in the diagnosis of lung illnesses. This will be especially beneficial in rural areas where radiologists are scarce. The dataset consisting of 1,12,000 chest X-ray images from the NIH was obtained from the Kaggle repository and used by the model. For predicting lung disease using chest X-ray images, The four deep learning models are incorporated in the proposed architecture namely, VGG16, ResNet50,EfficientNet B0, DenseNet 201 and then determined which model has the highest accuracy and performance. The learned model was tested using a dataset of NIH chest X-ray images from the repository of Kaggle. The full version of the dataset was taken into account. Tensorflow, Keras and Jupyter Notebook have been used for implementation. The model classified the test images as either healthy lung image or infected lung image. DenseNet 201 demonstrated accuracy of 71.5 percent, compared to accuracy of 69.6 percent, 69.1 percent, and 64.5 percent for VGG16, ResNet50, and EfficientNet B0, respectively. Considering different performance criteria such as recall, precision and validation accuracy, it was found that DenseNet 201 performed better than other approaches.