A CLAHE-Enhanced Vision Transformer with OVR-SVM for Breast Cancer Classification
S.M. Shahriar, Mitasree Barua, Nirzar Barua · 2025
Breast cancer refers to the uncontrolled cell mutation in the breast that often results in tumors. Ultrasound imaging is a reliable modality for detecting and evaluating breast lesions. However, its diagnostic accuracy is often constrained by subjective human interpretation and a limited field of view. To address these challenges, AI techniques, particularly machine learning (ML) and deep learning (DL) models, have been widely adopted. Among them, Vision Transformers (ViTs) have gained attention as a promising alternative for image classification, delivering both high accuracy and computational efficiency. In this study, the Vision Transformer (ViT) is used for deep feature extraction and combined with a One-vs-Rest Support Vector Machine (OVR-SVM) for classification, forming a robust hybrid approach. A curated set of the Breast Ultrasound Images (BUSI) was used, containing 789 images (445 benign, 211 malignant, and 133 normal) after filtering out low-quality and masked samples. The images were preprocessed using resizing, Contrast Limited Adaptive Histogram Equalization (CLAHE), and class balancing through oversampling to create 1000 images per class. The model was trained on 70% of the data, with 10% used for validation and the remaining 20% reserved for testing. It achieved a test accuracy of 98.83%, with 98.91% precision, 98.66% recall, and an F1-score of$\mathbf{9 8. 7 7 \%}$. Notably, the hybrid model outperforms traditional CNN-based methods and the standard ViT baseline, demonstrating superior generalization and predictive capability. This method offers an efficient and accurate tool for breast cancer diagnosis from ultrasound images, with strong potential for realworld clinical deployment.