Breast cancer detection using microscopic images based on composition features.

Breast cancer is one of the serious diseases that mainly affects women, and late diagnosis can lead to death. However, early detection significantly increases survival, thus making diagnosis very important. A reliable way to diagnose breast cancer is to analyze breast tissue samples under a microscope. Automatic classification techniques are very common in many fields in order to reduce human dependence. In this work, the combined process of three different methods for feature selection is used: (1) Speeded-Up Robust Features (SURF) to capture key structural and boundary points in microscopic images, (2) Local Binary Patterns (LBP) to extract local texture and intensity variations, and (3) statistical texture descriptors such as entropy, variance, contrast, and energy to quantify pixel-level distribution patterns. Our goal is to automatically determine whether a sample is malignant or benign. In the proposed method, the input dataset is divided into 5 categories, and cross-validation is performed with an accuracy of 95.89% (and an accuracy of 94.84% in the case of k = 10).