Abstract
INTRODUCTION: Breast cancer diagnosis in mammograms remains challenging due to limitations in preprocessing, accurate differentiation of benign and malignant cases, and precise tumor segmentation. METHODS: We propose Quantum-SpinalNet, a hybrid deep learning model combining Swin ResUNet3+ for tumor segmentation with a Deep Quantum Neural Network (DQNN) and SpinalNet for classification. Preprocessing involves CEAMF-based denoising, Z-score normalization, and context-aware contrast enhancement using spatial energy curves. Swin ResUNet3+ integrates ResUnet3+ decoders with Swin Transformer encoders for effective tumor localization and context extraction. RESULTS: Evaluation on the CBIS-DDSM and DDSM datasets demonstrates superior performance: accuracy 93.8%, sensitivity 94.1%, specificity 92.7%, precision 91.2%, F1 score 92.6%, Dice coefficient 0.89, and IoU 0.82. DISCUSSION: The proposed Quantum-SpinalNet provides a robust and interpretable framework for mammographic breast cancer detection, improving segmentation and classification precision, and supporting clinical diagnostic workflows.