Efficient Breast Cancer Classification with Transfer Learning and Ensemble Techniques for Imbalanced Data

Abstract

  Automated classification of breast cancer, particularly focusing on Invasive Ductal Carcinoma (IDC) detection, is crucial for timely and accurate diagnosis, significantly impacting patient outcomes. Leveraging deep learning techniques, this project aims to streamline histopathological image analysis for IDC identification, offering a cost-effective and efficient alternative to manual detection methods. The proposed approach involves deploying a lightweight ensemble model, combining a shallow CNN with MobileNetV2 via transfer learning, enhanced by data augmentation and hybrid techniques. Additionally, the project explores prediction techniques using Xception and DenseNet, contributing to further performance enhancement. Evaluation with diverse datasets demonstrates the efficacy of the ensemble model, with MobileNetV2 + Shallow CNN achieving 92% accuracy, while Xception exhibits superior performance with 95% accuracy. This advancement in automated histopathological image classification not only improves accuracy in IDC identification but also enhances overall screening effectiveness. Moreover, tailoring a lightweight ensemble CNN for edge devices addresses computational challenges, easing healthcare burdens and improving accessibility to quality care. The project signifies a significant step towards transforming breast cancer diagnosis, emphasizing the potential of deep learning in revolutionizing healthcare practices.