| Sumario: | Brain tumor classification plays a vital role in the medical diagnosis of brain lesions, ensuring accurate detection and supporting effective treatment planning. In this study, an ensemble learning framework is introduced based on a class-wise model selection strategy that integrates ResNet101 and DenseNet201 architectures to achieve enhanced classification accuracy. Both models were independently trained on a comprehensive brain MRI dataset containing 7041 images, using an 80/20 training–testing split. The ResNet101 model was trained on resized images, while the DenseNet201 model was trained on resized and CLAHE-preprocessed images. Additionally, the test dataset underwent image sharpening to improve structural details and boost classification performance. The proposed class-wise ensemble achieved an overall accuracy of 96.50%, precision of 96.43%, specificity of 98.84%, recall of 96.20%, and an F1 score of 96.25%, surpassing the performance of the individual models (ResNet101 and DenseNet201) across all evaluation metrics. These outcomes confirm the robustness and efficiency of the proposed framework, emphasizing the capability of class-wise ensemble methods to significantly improve brain tumor classification. This study contributes to the domain of medical image analysis by offering a reliable and highly precise approach for automated brain tumor diagnosis.
|
|---|