PulmoSight-XAI: An Explainable Multi-View Attention Ensemble with Gradient Boosting Meta-Learning for Multi-Label Chest X-Ray Classification

Multiple X-ray views improve diagnostic accuracy: The framework separately analyzes frontal and lateral chest X-rays rather than treating them identically, achieving state-of-the-art performance (0.9319 AUROC for frontal views), demonstrating that respecting the distinct information in different imaging angles leads to better AI-assisted diagnosis.

Addressing real-world medical challenges: The system tackles three practical problems that plague automated chest X-ray analysis—class imbalance (some diseases are rarer than others), co-occurring conditions (patients often have multiple pathologies simultaneously), and preservation of fine-grained details—making it more applicable to actual clinical scenarios.

Explainability is built-in, not added later: Rather than using a "black box" AI model and trying to explain it afterward, this framework incorporates explainability from the ground up through attention mechanisms that highlight which parts of the X-ray influenced each diagnosis, supporting clinician trust and verification.

Ensemble methods with intelligent combination: Instead of relying on a single neural network, the system combines five different models and uses advanced meta-learning techniques (gradient boosting) to intelligently weight their predictions, similar to how a hospital might consult multiple radiologists rather than one expert.

Designed for 14 simultaneous conditions: The framework classifies multiple thoracic pathologies at once (not just one disease per image), reflecting the complexity of real patient cases where several conditions may be present concurrently.