Classification and Interpretation of Histopathology Images: Leveraging Ensemble of EfficientNetV1 and EfficientNetV2 Models

Breast cancer is the second leading cause of cancer-related deaths among women, following lung cancer, as of 2024. Conventional cancer diagnosis relies on manual examination of biopsied tissues by pathologists which is a time-consuming process and based on pathologist experience may vary. Early detection and accurate diagnosis are critical for effective treatment planning and patient care. The invention of whole-slide scanners has revolutionized this process by enabling the adoption of Computer-Aided Detection (CAD) systems for automated analysis. Convolutional Neural Networks (CNNs) within CAD systems play a pivotal role in the automated classification of breast tissues. This study utilizes state-of-the-art CNNs called EfficientNetV1 and EfficientNetV2 for binary classification of BreakHis dataset ,a collection of histopathological images categorized as benign and malignant breast tissues. To address the challenge posed by the limited availability of annotated data, data augmentation and transfer learning techniques were applied. Model interpretability was enhanced using the Grad-CAM technique, which generates localization maps highlighting critical regions relevant to predictions. Finally, ensemble learning is employed for further improving performance. we utilized unweighted averaging and majority voting to combine predictions of multiple trained models. Furthermore, We define two ensemble architecture combining different trained EfficientNets. The proposed framework was able to achieve a classification accuracy of 99.58% outperforming conventional CNN models on BreakHis dataset. This study underscores the potential of ensemble learning to improve diagnostics accuracy in breast cancer detection.