Optimizing Breast Cancer Classification Accuracy with Hybrid Deep Learning and Advanced Image Processing

Breast cancer is one of the leading causes of mortality among women all over the world, specifically in regions where there is a lack of health care facility to detect the disease at an early stage. This study presents a machine learning (ML)-based approach for classifying breast cancer using histopathological images. Publicly available datasets such as the Wisconsin Breast Cancer Dataset (WBCD) and BreakHis are utilized. To improve the quality of images along with the performance of the model, intensive preprocessing procedures were adopted, i.e., data augmentation, image normalization, removal of noise, a histogram backtracking, and extracting the features were performed. Multiple ML models were implemented and evaluated, including Convolutional Neural Networks (CNNs), Random Forest (RF), Support Vector Machines (SVMs), and hybrid deep learning models. The highest classification accuracy (97.2%) was achieved by a hybrid model combining CNN with Local Binary Patterns (LBP) and Wavelet Transform, benefiting significantly from the image processing pipeline and class balancing methods. In order to take the model on supporting the model formation and estimation of the model performance, mathematical models that included cross-entropy loss, gradient descent, estimation metrics, including accuracy, precision, recall, F1-score and, the AUC-ROC were involved in this analysis. In as much as it offers, the study is limited by non-local data application and the DL models high training costs. Future studies need to focus on collecting localized data and integrating AI tools into the workflow of clinics to enable breast cancer detection earlier and more accurate in the setting of resource constraints.

Ҳали таржима қилинмаган