Performance Evaluation of MambaVision in Breast Cancer Detection from Histopathology Images

Authors : Hasan Zan

Pages : 879-888

Doi:10.24012/dumf.1691671

View : 57 | Download : 86

Publication Date : 2025-12-30

Article Type : Research Paper

Abstract :Accurate and early diagnosis of breast cancer is critical to improving patient outcomes, and histopathological image analysis remains the gold standard for clinical assessment. In this study, we explore the use of MambaVision, a hybrid visual state space and transformer model, for binary classification of breast histopathology images. Four pretrained MambaVision variants—Tiny, Small, Base, and Large—were fine-tuned and evaluated on a benchmark dataset to distinguish between benign and malignant tissue samples. The proposed models were trained using a two-stage learning rate schedule with cross-entropy loss, and performance was assessed using precision, recall, F1-score, and accuracy. Our experiments demonstrate consistently high classification performance across all model scales, with the Large variant achieving the highest accuracy of 99.7% and perfect recall. Confusion matrix analysis further highlights the models’ reliability in minimizing false negatives, a critical consideration in clinical applications. When compared to several existing deep learning approaches in the literature, MambaVision outperforms all competing methods, confirming its effectiveness in modeling both fine-grained cellular features and large-scale tissue context. The results suggest that MambaVision offers a scalable and accurate solution for computer-aided breast cancer diagnosis, with strong potential for deployment in digital pathology workflows.
Keywords : Meme Kanseri, MambaVision, Histopatolojik Görüntü Sınıflandırma, Derin Öğrenme

ORIGINAL ARTICLE URL