DETECTION AND PREDICTION OF BREAST CANCER USING AN ENSEMBLE DEEP LEARNING MODEL

ABSTRACT Breast cancer is one of the top causes of death among women worldwide. Early detection can greatly improve survival rates. However, existing computer-aided systems for diagnosis often struggle with issues like low image quality, poor performance across different datasets, and vulnerability to attacks that can reduce diagnostic accuracy. This study aims to create a reliable and accurate breast cancer detection system using a combination of advanced deep learning models and methods to defend against attacks. An ensemble model was developed by combining three convolutional neural networks: VGG16, ResNet50, and EfficientNetB0. A bagging approach with hard voting was used to improve accuracy. To strengthen the model, this study applied Projected Gradient Descent (PGD) for adversarial training. The study also used several datasets, including INBreast, CBIS-DDSM, and Breast MRI, and applied preprocessing techniques like contrast enhancement (CLAHE), wavelet transformation, and edge detection to improve image quality. Additionally, a symptom-checking tool was included that uses logistic regression to give real-time risk evaluations based on user health data. The ensemble model’s performance was evaluated based on accuracy, precision, recall, specificity, F1-score, and AUC-ROC, which are 86.12%, 80.27%, 82.07%, 88.44%, 81.16%, and 93.98% respectively. This study showed that the ensemble model performed better than individual models in accuracy, recall, and resistance to attacks. The system also provides a user-friendly interface for uploading images and receiving diagnoses. This study advances the field by combining reliable deep learning with various data types and a practical design, making breast cancer detection more dependable and accessible, especially in areas with fewer resources. Keywords: Breast Cancer Detection, Ensemble Learning, Adversarial Training, Medical Imaging, Deep Learning Robustness