Enhancing Breast Cancer Detection Using Multistage Transformer with Positional Encoding and Feature Fusion

Breast cancer (BC) remains the most common cancer among women worldwide. However, early detection and treatment have significantly improved the disease's prognosis and decreased mortality rates. Breast abnormalities are detected either through a physical examination or an automatic screening method. Pathologists examine breast pictures to detect tumors; however, it is time-consuming and requires a high degree of knowledge and experience. Thus, an automated image analysis fastens the process and increases the accuracy of diagnosis. To identify BC, various machine learning (ML) and deep learning (DL) approaches have been utilized in the past. However, the complicated and varied nature of breast tissue makes BC diagnosis challenging. Convolutional Neural Networks (CNNs) are frequently used for breast image classification; however, their performance at different magnification levels is constrained by their fixed-size kernels and small receptive fields. To tackle these constraints, we propose a novel Multi-Stage Transformer with Positional Encoding and Feature Fusion (MST-PEFF). The local feature extraction through CNN, along with transformer-based global modeling, makes it effective for histopathological image classification. Based on experimental results, the proposed model achieved 94.7% classification accuracy on the BreakHis dataset, outperforming baseline methods.