Quantum-Enhanced Transfer Learning for MS Lesion Detection in MRI: A Hybrid Classical-Quantum Framework

Accurate detection of multiple sclerosis (MS) lesions in magnetic resonance imaging (MRI) requires robust deep learning models to capture subtle spatial and textural features. We introduce hybrid quantum-classical transfer learning algorithms for MS classification using axial and sagittal MRI scans, combining classical convolutional neural networks (CNNs) including EfficientNetB3, ResNet50, DenseNet121 with parameterized quantum circuits to enhance feature representation via entanglement and quantum-specific non-linearities. Quantum layers are trained end-to-end with classical backbones via backpropagation, enabling seamless integration of quantum-enhanced features. For axial MRI, QResNet50 achieved a high accuracy of 97.58% and AUC of 99.31%, while QDenseNet121 reached 97.28% accuracy and 99.13% AUC. For sagittal MRI, classical ResNet50 excelled with 99.15% accuracy and 99.93% AUC, while QEfficientNetB3 improved accuracy (97.46% to 98.30%) but reduced AUC (99.51% to 99.32%), and QDenseNet121 achieved 98.87% accuracy and 99.83% AUC. Hybrid models showed mixed results, with QCNN underperforming, suggesting quantum benefits are architecture-dependent. Despite simulated quantum circuits mitigating hardware limitations, our results demonstrate the potential to enhance diagnostic performance in specific architectures. This work clarifies a foundational step toward quantum-enhanced deep learning for clinical applications, opening research directions in quantum-aware transfer learning and error mitigation for biomedical imaging.