Breast cancer classification in ultrasound imaging using Mamba: An efficient state-space model for computer-aided diagnosis

Abstract

Breast cancer remains a leading cause of cancer-related deaths among women world- wide. Ultrasound imaging provides a non-invasive, low-cost diagnostic modality, but its interpretation is prone to variability. This study proposes a novel Mamba-based state-space deep learning model for the classification of breast ultrasound images into normal, benign, and malignant categories. A total of 2400 ultrasound images were preprocessed using CLAHE and patch-based tokenization, then classified using a Mamba-based state-space model with optimized hyperparameters. Performance was assessed using k-fold cross-validation, and results were statistically compared with CNN and Transformer-based baselines. The proposed model achieved an accuracy of 91.1%, sensitivity of 90.6%, specificity of 91.3%, F1-score of 90.9%, and AUC of 0.945, demonstrating comparable performance with reduced computational cost. The proposed approach demonstrates competitive performance and computational efficiency, highlighting its potential as a research-oriented computer-aided analysis tool for breast ultrasound imaging.