Automated Pneumonia Diagnosis from Chest X-rays Using Deep Learning Techniques

Pneumonia is a common and potentially life-threatening respiratory infection. Accurate and timely diagnosis is critical for effective treatment, but traditional diagnostic methods can be time consuming and require specialized expertise. Deep learning, a subset of artificial intelligence, has shown promise in improving pneumonia diagnosis through analysis of medical imaging. The novelty of our research stems from the unique approach we used to collect our dataset of normal and pneumonia patients. Rather than relying on pre-existing datasets, we employed a novel data scraping method to collect data from Bing images. By utilizing this novel data collection method, we were able to overcome limitations of existing datasets, such as potential biases and limited sample sizes, and obtain a more representative and robust dataset for our study. Our dataset, thus, offers a unique and valuable resource for studying the relationship between pneumonia and various patient factors, providing new insights and opportunities for research in this field. This innovative data collection approach sets our study apart and contributes to the novelty and rigor of our research findings. In this study, we trained and evaluated five deep learning models (CNN, VGG16, ResNet50, DenseNet169 and InceptionV3) on a dataset of chest X-ray images for pneumonia diagnosis. We found that all models achieved high accuracy and demonstrated strong precision and recall values. The VGG16, ResNet50, and DenseNet169 models showed promising results with testing accuracies of 90%, 91%, and 90%, respectively. The InceptionV3 model had a lower testing accuracy of 89%. The precision-recall curve analysis showed high AUC values for all models, indicating good performance in detecting pneumonia. Our findings suggest that deep learning models can be effective tools for pneumonia diagnosis, with the potential to improve accuracy and efficiency of traditional diagnostic methods. However, further research is needed to address limitations and challenges, such as potential bias in dataset selection and limitations in generalizability. With careful consideration and validation, deep learning models have the potential to play a valuable role in improving pneumonia diagnosis and ultimately, patient outcomes.