Breast Cancer Classification with Image Processing and Deep Learning Models

Early detection of breast cancer is a critical global health challenge, and accurate classification of ultrasound images plays a vital role in diagnosis. This study presents an effective deep learning-based approach for detecting breast cancer using grayscale ultrasound images. We evaluated five different models, including popular deep learning architectures such as DenseNet121, VGG19, VGG16, AlexNet, and a hybrid method combining ResNet50V2 feature extraction with an XGBoost classifier, on a publicly available ultrasound image dataset. Among these, DenseNet121 achieved the highest accuracy at 99.18%, followed by VGG16 (98.70%), VGG19 (98.15%), AlexNet (98.11%), and the ResNet50V2 + XGBoost method (94.92%). Additionally, we developed a custom convolutional neural network (CNN) that achieved 98.18% accuracy, demonstrating competitive performance. The novelty of this work lies in the comprehensive comparison of multiple pre-trained architectures alongside a custom CNN and a hybrid feature-based approach, all applied to the same ultrasound dataset-offering a unified benchmark for breast cancer classification. The results demonstrate that deep learning models, particularly DenseNet121, can significantly improve the accuracy and reliability of ultrasound-based breast cancer diagnosis.