DenseNet201: A Deep Learning Method for Improving Pulmonary Emphysema Identification from Chest X-rays

Lung diseases are the leading causes of early death and disability worldwide. Each year, they kill 4 million people and make them unable to work. One of the pulmonary diseases is emphysema. Emphysema, caused by the breakdown of alveolar walls and a lack of elasticity, is one of the most prevalent diseases responsible for this condition. Many algorithms have been developed in the past to classify emphysema. The objective of this article is to create a transfer learning-based model to determine whether a patient has emphysema based on chest X-rays. DenseNet201, a CNN classifier based on transfer learning, is proposed in this research. The suggested approach was tested using the NIH chest X-ray dataset (7540 samples). Finally, the experiments revealed that the DenseNet201 has a classification accuracy of 98.87%, an AUC of 99.7%, an F1-score of 98.88%, a precision of 97.80%, an AUC of 99.7%, a sensitivity of 100%, a precision of 97.80%, a validation accuracy of 98%, and a specificity of 97.75%, which is higher than other models. Other hyperparameters, such as learning rates, epochs, batch size, number of filters, and activation function, have been tuned repeatedly to improve results. The main goal is to have a low number of false positives and a 0% false negative rate.