Fusion Learning Framework for Malignant and Benign Classification in Histopathology Images

This work presents a robust ensemble-based deep learning approach for classifying breast cancer histopathology images from the BreakHis dataset into Malignant and Benign categories. This study implements three ensemble strategies efficientNetB0 and ResNet50, EfficientNetB0 and VGG16, and VGG16 and ResNet50―the study highlights the potential of combining complementary deep learning architectures for improved classification performance. Each ensemble incorporates modifications to optimize feature extraction and generalization for the specific dataset, including dense layers, dropout, and feature concatenation techniques. The proposed methodology achieves high accuracy, precision, recall, and F1-scores across all methods, with the EfficientNetB0 and VGG16 ensemble demonstrating superior performance. Comparative analysis with state-of-the-art methods with robustness and adaptability of the proposed approach, offering a scalable and efficient framework for medical image classification. This work provides valuable insights into the application of ensemble learning for reliable cancer diagnosis.