Attention-Based Decision Fusion for Breast Cancer Classification Using Ensemble Transformers

Abstract

Breast cancer is a significant global health issue, and early diagnosis and accurate diagnostic processes are of great importance. In this study, deep learning-based transformer models were fine-tuned using a transfer learning approach for the classification of breast cancer histopathological images, and an attention mechanism-based decision fusion method was proposed to optimize model predictions. Experiments conducted on a widely used dataset in the literature demonstrated that the highest classification performance among individual models was achieved with an accuracy rate of 92.25%. However, using the proposed attention-based fusion method, an accuracy rate of 95% was attained on the test set. Additionally, analyses performed on an independent hidden test dataset to evaluate the model's generalization capability achieved an accuracy rate of 90%, indicating that the proposed method provides an effective solution for breast cancer classification.