Search

Breast cancer remains a leading cause of cancer-related mortality among women worldwide, underscoring the importance of effective screening strategies for early detection and improved survival. Although conventional modalities such as mammography reduce mortality, they are limited by false positives, false negatives, and overdiagnosis, particularly in dense breast tissue and diverse populations. Deep learning, especially convolutional neural networks (CNNs), has shown promise in improving diagnostic accuracy and reducing inter-reader variability; however, its translation into routine clinical practice requires critical evaluation beyond reported performance metrics. This critical review evaluates CNN-based deep learning applications for breast cancer detection across mammography, ultrasound, and MRI, with emphasis on training strategies and barriers to clinical deployment. A targeted literature search identified peer-reviewed studies focusing on CNN architectures, transfer learning, and implementation challenges. Findings indicate that models such as ResNet, DenseNet, and EfficientNet perform well in controlled settings, supported by transfer learning and data augmentation approaches. However, these results often fail to translate into consistent clinical performance, particularly across imaging modalities and real-world workflows. Limitations including demographic bias, insufficient external validation, and weak evidence of outcome or cost-effectiveness highlight a substantial gap between experimental success and clinical readiness. The review concludes that while deep learning in breast imaging is promising, its adoption should remain cautious and evidence-driven until robust clinical benefit is clearly demonstrated.

Neoadjuvant chemotherapy (NAC) is standard for locally advanced breast cancer, with pathologic complete response (pCR) strongly predicting improved survival. However, only 30–40% of patients achieve pCR, while the rest undergo toxicity and delayed surgery without benefit. Current prediction methods rely on tumor volume at isolated time points or simple pre- and post-treatment comparisons, ignoring continuous tumor dynamics during therapy. Sparse and irregular MRI sampling further limits accurate modeling. We introduce a Neural Ordinary Differential Equation (Neural ODE) framework to model continuous tumor growth from sparse serial MRI during NAC. The model learns a time-continuous function describing tumor evolution and predicts individual response trajectories and final pCR status. The framework includes (1) MRI-based tumor segmentation, (2) construction of sparse longitudinal tumor volume series, (3) Neural ODE modeling of continuous dynamics via a neural network–parameterized derivative function, and (4) classification of the final latent state for pCR prediction. An optional module enables trajectory visualization and interpretability. This approach captures hidden continuous tumor behavior between scans, handles irregular sampling without imputation, and enables earlier response prediction. It is also computationally efficient using adjoint-based training and may reveal distinct growth patterns between responders and non-responders. Neural ODE-based modeling offers a more informative framework for predicting NAC response by capturing continuous tumor dynamics, with potential to improve pCR prediction over conventional volume-based methods.