Explainable Cost-Sensitive Deep Neural Networks for Brain Tumor Detection from Brain MRI Images considering Data Imbalance

Explainability matters for medical AI: The researchers incorporated Explainable AI techniques to visualize how the models make decisions, addressing the "black box" problem that can make doctors hesitant to trust automated systems in clinical settings.

Handling imbalanced data improves real-world performance: Cost-sensitive neural networks achieved nearly 4% higher accuracy than conventional models when working with imbalanced datasets, which is crucial since real medical data often has unequal numbers of tumor vs. non-tumor cases.

Perfect recall can be more important than perfect accuracy in medical contexts: The cost-sensitive models achieved 100% recall (catching all actual tumors), which is critical for cancer detection—missing a tumor is more harmful than a false alarm, even if overall accuracy is slightly lower at 92.31%.

Multiple model architectures should be tested: The study compared five different deep learning models (CNN, ResNet50, InceptionV3, EfficientNetB0, and NASNetMobile), finding that fine-tuned InceptionV3 performed best at 99.33% accuracy on balanced data, suggesting no single approach works universally.

Transparency and reproducibility strengthen responsible AI: The authors made their datasets and code publicly available, enabling independent verification and building trust in the system—an important practice for medical AI tools that will affect patient care.