An Ensemble Learning Approach using Self-Supervised and Meta-Learning for Few-Shot Pneumonia Detection in Chest X-Ray Images

Pneumonia remains one of the leading causes of mortality worldwide, particularly among children and the elderly. Early detection is critical for improving patient outcomes; however, traditional diagnostic methods require expert radiologists and large amounts of labeled data, often scarce in resource-limited settings. To address this challenge, this study proposes SimCLR-MAML, a hybrid model that integrates self-supervised contrastive learning (SimCLR) with model-agnostic meta-learning (MAML) to enhance few-shot pneumonia detection from chest X-ray images. The SimCLR module enables effective feature extraction from unlabeled images, while MAML facilitates rapid adaptation to new classification tasks with limited labeled data. The proposed SimCLR-MAML model was evaluated using standard performance metrics, including accuracy, precision, recall, F1-score, and AUC (Area Under the Curve). The experimental results demonstrated that SimCLR-MAML achieved an accuracy of 93.21%, with a precision of 93.42%, a recall of 93.21%, and an F1-score of 93.21%, significantly outperforming MAML (76.00% accuracy) and demonstrating comparable performance to SimCLR (93.69% accuracy). The hybrid model balances feature extraction and adaptability, making it a robust solution for automated pneumonia classification with limited annotated data. Despite its promising results, the study identifies areas for further research, including enhancing interpretability through explainable AI techniques, optimizing the model for real-time deployment in low-resource settings, and validating its generalizability across multiple datasets and healthcare institutions. The findings highlight the potential of combining self-supervised learning and meta-learning to bridge the gap between data scarcity and effective deep learning-based disease detection.