Customized VGG16 Model for Histopathological Image Classification with Class Imbalance Solutions

Breast cancer remains a global health issue and the need for early and robust detection is immense. The best approach to detect breast cancer is histopathological imaging, yet hand-based assessments can be arbitrary and differ depending on the observer. Recent advancements of Deep learning architectures revolutionize automatic classification of breast cancer types from histopathological images. This paper presents a modified VGG16-based CNN model for multiple group sorting of breast cancer histological images. The proposed approach employs transfer learning, the Synthetic Minority Over-Sampling Technique (SMote), and data augmentation to fix the class imbalance problem and enhance model generalization. Early Stopping, ReduceLROnPlateau, and dropout regularization, among other optimization techniques are employed to ensure performance stability. The experiments on the publicly available BreaKHis dataset achieves a classification accuracy of 96.9%. The model outperforms previous reported works in terms of precision, recall, F1-score, and confusion matrix performance measurements. The proposed approach serves as a robust and accurate model for breast cancer classification by addressing fundamental challenges in autonomous histopathology image analysis.