Deep Dynamic Ensemble Classification for Brain Tumor MRI

Traditional models for brain tumor diagnosis often struggle with limited prediction accuracy, largely due to their inability to fully capture the heterogeneous nature and complexity of tumor characteristics. Moreover, they lack the adaptability required for generalization across diverse datasets. In response to these limitations, this paper proposes a novel brain tumor classification framework using dynamic ensemble learning to improve diagnostic accuracy. The proposed framework combines a pool of pre-trained convolutional neural networks (CNNs) with an adaptive selection mechanism to dynamically choose the most effective classifier based on task-specific metrics. This dynamic adjustment of CNN weights integrates each model’s strengths, boosting predictive accuracy and robustness. Extensive experiments across various datasets confirm our approach outperforms single models and the existing methods on multiple performance indicators