Deep Learning Approach for Breast Cancer Image Classification using Multi-Modal Image

Abstract

Artificial Intelligence (AI) plays an important role in breast cancer detection, helping radiologists quickly and accurately identify and classify the disease. Among AI methods, deep learning models like Convolutional Neural Networks (CNNs) perform well in medical image analysis. This paper proposes a deep learning architecture for breast cancer classification, evaluated on two imaging modalities, mammograms (INBreast) and ultrasounds (BUS dataset). The model integrates depthwise separable convolutions (DSConv) and Squeeze-and-Excitation (S&E) blocks to efficiently capture both low and high-level features while focusing on the most relevant spatial and channel-wise information. Preprocessing and data augmentation techniques are applied to improve generalization and reduce overfitting. Hyperparameters, including optimizers and epochs, are systematically tuned. The performance is evaluated and achieves accuracies of 98.79% and 99.28% on INBreast and BUS datasets, respectively. These results demonstrate strong performance across different imaging types and highlight the model’s potential for clinical application.