Unsupervised Deep Feature Extraction For Pulmonary Disease Detection Using CT And X-Ray Imaging

Abstract

Lung disorders cause major mortality worldwide. Early diagnosis improves recovery and long-term survival, making this task urgent. The research introduces Lung-GANs, a deep unsupervised framework. A Generative Adversarial Network (GAN) deep learning model learns lung disease picture representations from unlabeled data in this context. The unsupervised framework permits learning without labeled instances, which is useful. Lung-GANs help clinicians detect lung problems quickly, accurately, and automatically. Automated detection can speed up diagnosis and treatment. Lung-GANs learn interpretable lung disease images. The GAN algorithm extracts relevant visual features to better understand lung disease trends. SVMs and voting classifiers are trained using Lung-GAN features. This feature extraction and categorization method is novel. Voting classifiers use many models' predictions to create a robust classification framework, while SVMs excel at binary classification. The advanced YOLOv5 and YOLOv8 methods obtain 99% mAP for CT-scan pictures, improving object recognition. Flask-based front ends make CT and X-ray image testing straightforward, increasing user engagement. Integration of authentication provides a complete system security solution for real-world applications.