Deep Learning-Based Diagnosis of Pneumonia Using Convolutional Neural Networks

Abstract

Pneumonia is a respiratory illness characterized by lung inflammation, often caused by pathogens such as viruses, bacteria, or fungi. Timely detection of pneumonia is crucial for effective treatment. While chest X-rays are commonly used for diagnosis, manual interpretation can be time-consuming, particularly in areas with limited access to trained radiologists. Currently, deep learning models have emerged as an efficient method for pneumonia diagnosis. Numerous researchers are dedicated to enhancing pneumonia diagnostic capabilities through artificial intelligence methods. This study employs a convolutional neural network (CNN) for pneumonia diagnosis. The dataset used in this study consists of chest X-ray images of healthy individuals as well as those affected by bacterial and viral pneumonia. In this study, a CNN model is implemented using an imbalanced chest X-ray dataset with a weighted cross-entropy cost function. The outcome of the developed CNN model shows an accuracy of 75.84%, a precision of 83.16%, a recall of 68.37%, and an F1 score of 68.97% on the test dataset. Further tuning of the model’s hyperparameters is necessary to improve performance metrics.