A Comparative Performance Evaluation of Classical and Quantum-Based Deep Learning Models in the Classification of Breast Cancer Histopathological Images

Breast cancer histopathological image classification is an important task for computer-aided diagnosis, yet patch-level analysis remains challenging due to tissue heterogeneity, visual similarity between classes, and the risk of patient-level data leakage. This study evaluates classical and quantum-assisted deep learning configurations for binary invasive ductal carcinoma classification using histopathological image patches. Six models were compared using a common patient-disjoint split: a task-specific convolutional neural network, ResNet18 and DenseNet121 transfer learning models, a hybrid quantum convolutional neural network, and two quantum transfer learning models based on ResNet18 and DenseNet121 backbones. The models were assessed using accuracy, precision, recall, F1 score, ROC-AUC, and PR-AUC. The classical convolutional neural network achieved the best overall performance, with 88.00% accuracy, 90.00% recall, 0.9624 ROC-AUC, and 0.9677 PR-AUC. Among the quantum-assisted configurations, QTL-DenseNet121 achieved the strongest result, with 86.50% accuracy and 0.9315 ROC-AUC, while the hybrid quantum convolutional model achieved 81.00% accuracy using 144 trainable quantum parameters. The findings indicate that compact quantum-assisted classifier components can be feasibly integrated into medical image classification pipelines, although the results do not demonstrate quantum advantage and require further validation.