A Novel Ensemble Model Integrating Efficient Net for Improved Diagnosis in Medical Imaging

Medical image classification plays a crucial role in early disease diagnosis and
treatment planning. However, achieving high accuracy and robustness remains a challenge due
to variations in imaging modalities, noise, and dataset limitations. In this study, we propose a
hybrid ensemble model integrating EfficientNet with ResNet, Xception, and DenseNet to
improve classification performance in medical imaging. The proposed framework leverages
transfer learning and feature fusion techniques, where deep feature embeddings from multiple
CNN architectures are combined and processed using a meta-learner (XGBoost or Logistic
Regression) for final classification. The model is trained and evaluated on benchmark medical
imaging datasets, including brain tumor MRI, chest X-ray pneumonia, and retinal disease
classification.
Experimental results demonstrate that the ensemble approach significantly outperforms
individual models, achieving higher accuracy, robustness, and generalization compared to single deep learning architectures. Furthermore, we employ attention mechanisms and Principal Component Analysis (PCA) for optimal feature selection, reducing redundant information while maintaining high diagnostic performance. The proposed approach offers a
promising solution for real-world medical image analysis, enhancing automated disease detection with improved precision and reliability