Early Detection of Breast Cancer through Automated Radiology Image Classification using Machine Learning Model

Background: Breast cancer is still a big concern all around the world, and finding it early greatly improves life expectancy for patients. Methodology: This study presents an automated breast cancer detection approach using the ResNet50V2 model on histopathology images. The dataset comprises 278,082 images categorized into Invasive Ductal Carcinoma (IDC) and Non-IDC classes. Preprocessing steps include grayscale conversion, Contrast Limited Adaptive Histogram Equalization (CLAHE) for enhanced contrast, image resizing, and Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. The ResNet50V2 model, enhanced with additional dense layers, undergoes fine-tuning and is trained using categorical cross-entropy loss. Results: The model achieves 87.49% accuracy on the test set, according to performance evaluation utilizing accuracy, precision, recall, F1-score, confusion matrix, and ROC curve. Conclusion: A comparative analysis with VGG16 and DenseNet121 confirms the superior classification ability and generalization of ResNet50V2. These findings highlight deep learning’s potential for clinical applications, offering a reliable method for early breast cancer diagnosis.