Noise-Resilient, Explainable Classifications of High-Resolution Images Using Quanvolution-Pooling

Applying Quantum Machine Learning (QML) to high-resolution image classification is a significant challenge, typically requiring extensive classical pre-processing. This work proposes a QML model that uses a novel quanvolution-pooling to directly extract features from high-resolution images, eliminating the need for classical pre-processing with fixed or learnable operations. By utilizing a compact configuration of only hardware-native gates, our approach achieves a high degree of inherent noise resilience. We also present the first LIME-based explainability study of QML models for high-resolution image classification tasks. Our approach was evaluated on three datasets under both noise-free and noisy simulations, exhibiting minimal performance degradation in the presence of noise. Explainability analysis revealed that the model consistently focused on clinically relevant regions, even under noisy conditions. These findings demonstrate the feasibility and robustness of QML models for high-resolution image classification.