I'm exploring optimal CNN architectures for breast cancer detection using deep learning. I want insights into effective feature selection techniques and hyperparameter tuning strategies to enhance model performance and generalization. What are the best approaches for fine-tuning pre-trained models for medical imaging?