Research on Classification Methods of Breast Cancer Based on Deep Learning

This research presents a novel system that combines MobileNetV2 structure, attention systems, and a sophisticated feature pyramid network to categorize breast cancer X-ray images into three distinct groups: normal, benign, and malignant. Utilizing the Breast Cancer Dataset, the dataset was meticulously divided into training, validation, and testing segments, keeping the ratio at 8:1:1. To improve feature extraction during the training, we employed preprocessing methods such as random cropping, horizontal flipping, the addition of Gaussian noise, and the variation of colors. The findings indicate that the proposed MobileNet-based model exhibits strong classification skills, rapid convergence, achieving 98.56% accuracy, and a negligible loss rate of 0.075. The brief and precise method employed is vital for accurately classifying breast cancer X-ray images.