GAN augmented framework with hybrid deep learning techniques for addressing data scarcity in medical image analysis

Chest X-ray analysis plays a critical role in screening and stratifying COVID-19 and pneumonia patients, but real-world datasets often suffer from class imbalance and limited high-quality samples. This creates challenges for deep learning models, particularly in achieving reliable and generalizable performance across diverse clinical settings. We propose F-CoVaD, a GAN-augmented deep ensemble framework that integrates class-specific synthetic image generation with complementary feature learners. The pipeline incorporates DCGAN and Vanilla GAN for balanced data augmentation, CNN models (VGG16 and ResNet50) for hierarchical spatial feature extraction, and an LSTM module for modeling long-range spatial dependencies within the feature maps. Ensemble fusion is performed using validation-driven weighting to enhance robustness, while explainable AI techniques are included to provide interpretable, clinically meaningful visualizations of model decisions. The unified design of generative augmentation, multi-architecture feature learning, and explainability forms a robust approach for CXR-based disease classification. The framework demonstrates improved stability under class imbalance and offers transparent and clinically aligned diagnostic insights. This work is novel in attempting an unique integration of class-specific GAN augmentation, a hybrid CNN–LSTM ensemble, and statistical validation into a single, generalizable pipeline for CXR-based COVID-19 diagnosis.