A hybrid feature extraction based ensemble model for breast cancer detection and classification using different medical images

Breast cancer is a fatal disease that has a high death rate worldwide, according to the WHO. Hence, implementing medical image-based automated breast cancer detection and classification is essential for early identification and categorization. It plays a crucial role in developing efficient treatment methods by accurately diagnosing the kind and classifying the subtype of breast cancer. Ultrasound and mammograms are primary and efficient methods for detection, whereas histopathology is an advanced method for exactly classifying breast cancer. Previously, different hand-engineered features were used for different types of data sets, respectively, which provided high accuracy individually. However, deep learning is a strong tool for computer vision tasks. Therefore, we developed a unique combination of hand-engineered features for color, shape, and texture extraction in parallel to three different deep neural networks. Such a hybrid method proposed that combines both hand-engineered and deep learning-based feature extractors provides an outstanding performance for breast cancer detection and classification on different types of datasets compared to the state of the art methods thus verifying its robustness and effectiveness. • A hybrid feature extraction method that combines handcrafted features with deep learning. • A unique combination of handcrafted features is established by focusing on color, shape, and texture features for ultrasound, mammography, and histopathology images. • Ensemble network integrating transfer learning and handcrafted features with deep neural network (DNN) for breast cancer detection and classification using different medical images.