Multi-dataset analysis demonstrates that deep learning models outperform classical methods in breast cancer detection, suggesting better integration into clinical practice.
Key Points
Deep convolutional neural networks like ResNet-18 achieved an accuracy of 99.7% in breast cancer detection.
Grad-CAM visualizations improved transparency of the models by highlighting diagnostic features in ultrasound images.
Classical machine learning models can still provide competitive performance when enhanced with deep feature extraction.
The study integrates multiple datasets, showcasing the feasibility of AI-driven diagnostic tools in clinical workflows.