Real-Time Explainable Pulmonary Disease Detection Using MAE-Enhanced Swin Transformer Feature Extraction with Faster R-CNN on Chest X-Ray and CT Scan Images

Abstract

A proposed framework for real-time and explainable pulmonary disease detection is based on chest X-ray images and computed tomography (CT) images. It is constructed from a Masked Autoencoder (MAE) and a Swin Transformer by incorporating them into one model to obtain robust multi-scale features. The MAE can be used for self-supervised learning to benefit from noise-resilient feature learning, and the Swin Transformer can be used to capture the local and global contextual dependency. Features are extracted and passed to Faster R-CNN network for precise localization and classification of pulmonary abnormalities. Proactive use of Explainable Artificial Intelligence (XAI) techniques for enhancing interpretability and clinical trust. A hybrid optimization strategy is further used to improve the detection and localization performance. Proposed MAE-ST-FRCNN model provides accuracy of 92.7%, Precision of 92.0%, Recall of 91.6% and F1-score of 85.1%. It also achieves a high AUC value of 96.2%, which shows high discriminative power. The model attained a score of 85.1% on Jaccard Index, which has provided accurate localization of abnormal regions.