Enhancing breast cancer diagnosis with ResNet50 and Salp Swarm-Based feature reduction on BreakHis dataset

Background Histological analysis remains the gold standard for breast cancer diagnosis, with patient survival rates depending heavily on the precise interpretation of pathological images. In recent years, the integration of computer-aided diagnosis (CAD) systems with pathology imaging has shown great potential for improving the accuracy and efficiency of physicians’ assessments. Such automated systems not only reduce diagnostic time but also help in detecting and monitoring breast tumors more effectively. Objective This study aims to develop an automated and highly accurate system for classifying histological images of breast cancer using advanced artificial intelligence techniques. The ultimate goal is to enhance diagnostic speed and reliability while reducing the workload on physicians, thereby contributing to improved patient care and outcomes. Method The proposed approach employs the ResNet50 deep learning architecture for feature extraction from histological images in the BreakHis dataset. To optimize performance, the Salp Swarm Algorithm (SSA) is applied to reduce the dimensionality of the extracted features. These optimized features are then used to train four machine learning classifiers: Decision Trees (DT), K-Nearest Neighbors (KNN), Random Forests (RF), and Support Vector Machines (SVM). Performance is evaluated across different image magnifications to ensure robustness. Results Among the tested models, the SVM classifier consistently achieved the highest accuracy, recording 96.35 % at 40 × magnification, 94.10 % at 100×, 96.40 % at 200×, 96.09 % at 400×, and 96.45 % on the aggregated dataset. These results demonstrate the model’s strong generalization capabilities across varying magnifications of histological images. Conclusion The results show that by facilitating the quick and precise identification of breast tumors from histological images, the suggested approach can significantly reduce the time required for manual diagnosis. This capability is critical in oncology, where timely detection directly influences treatment outcomes. By combining deep learning-based feature extraction with optimization-driven dimensionality reduction, the system offers an effective solution for improving diagnostic precision, efficiency, and patient survival rates.