Deep Learning for Anomaly Detection and Classification of Lung Images

Abstract

Lung diseases such as pneumonia, TB, COVID-19, and lung cancer have a huge influence on global healthcare systems and human well-being. For early diagnosis and better patient outcomes, these disorders must be identified and categorized early. Lung X-ray images are mostly used for examining and identifying these diseases due to their non-invasiveness and cost-effectiveness. In this work, deep learning approaches are created to improve the detection of anomalies and disease classification in lung X-ray images. The work initially focused on disease classification using CNN, where the model was trained to accurately classify images into different lung diseases, such as pneumonia, tuberculosis, COVID-19, and lung cancer. This allowed us to provide specific disease labels to X-ray images, aiding in effective treatment planning. To further enhance our system's capabilities, the work was expanded to incorporate anomaly detection. By leveraging the power of Mask R-CNN, a separate module to detect subtle abnormalities and early signs of disease progression in lung X-ray images is developed. This enabled us to perform a more in-depth inspection of the images by allowing us to comprehend abnormalities. By utilizing publicly available datasets containing negative and positive instances of various lung diseases, the proposed work extensively assessed the performance and accuracy of the detection and classification models. The results showed significant improvements in both disease classification and anomaly detection, indicating the potential of our approach in assisting medical professionals with timely and accurate diagnoses. The proposed work provides a flexible and modular strategy to improve lung X-ray image analysis.