Ensemble deep learning framework integrating deep image features and statistical descriptors for robust tumor diagnosis
Yunus Emre Göktepe · Biomedical Signal Processing and Control · 2026
Since breast cancer is still the world’s leading cause of death for women, reliable and accurate diagnostic systems must be developed. In order to classify breast cancer, we present EBCNet, a hybrid ensemble deep learning architecture that combines hand-crafted statistical descriptors with deep features based on Convolutional Neural Networks (CNNs) in a synergistic manner. The framework’s adaptability to a variety of data types is demonstrated by its validation on both structured (WBCD) and image-based (MIAS) datasets. Three manually designed features—Robust Median Deviation (RMD), Wavelet Energy, and GLCM Contrast—are utilized to improve feature representation, and a lightweight SqueezeNet model is utilized to extract rich deep features from mammography images. Dense layers are used to further transform these features, and an ensemble of SVM and XGBoost classifiers is used for classification. According to experimental results, EBCNet achieves 99.30% accuracy on WBCD and 96.55% accuracy on MIAS, outperforming or matching the state-of-the-art models. The effectiveness and robustness of the suggested hybrid approach are confirmed by an ablation study, which also validates the individual contributions of the handcrafted features. In addition to offering promising potential for wider applications in computer-aided medical decision systems, this work offers a useful, interpretable, and generalizable solution for the diagnosis of breast cancer.