Mohammad Zahaby; Mostafa Boroumandzadeh; Iman Makhdoom
Abstract
Breast cancer is one of the most prevalent cancers among women and remains a leading cause of cancer-related mortality. Mammography is the primary imaging modality for the early detection of breast tumors. Providing timely and highly accurate diagnoses is a top priority for physicians ...
Read More
Breast cancer is one of the most prevalent cancers among women and remains a leading cause of cancer-related mortality. Mammography is the primary imaging modality for the early detection of breast tumors. Providing timely and highly accurate diagnoses is a top priority for physicians and healthcare providers in the management of critical illnesses. This paper presents a Medical Decision Support System (MDSS) that utilizes Yager’s rule of combination to classify and diagnose breast cancer patients by integrating information from multiple data sources. Medical text reports (MTR) and key feature vectors extracted from electronic health records (EHR) were reduced using Principal Component Analysis (PCA) and then classified using Convolutional Neural Networks (CNN), Multi-Layer Perceptrons (MLP), and Support Vector Machines (SVM). Medical images were preprocessed and classified using a U-Net model. A novel decision fusion algorithm, called weighted Yager, was introduced to determine the Breast Imaging-Reporting and Data System (BI-RADS) categories, taking into account the accuracy of each class in each classifier as evidence. The performance of the proposed system was evaluated based on standard metrics including accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1-score. The proposed system achieved the highest accuracy of 96.23\%, outperforming individual classifiers (CNN: 86.37%, MLP: 92.11%, SVM: 87.92%, U-Net: 92.97%, and Yager: 93.49%). The weighted Yager fusion method yielded the best performance with an accuracy of 96.23%, sensitivity of 98.80%, specificity of 85.90%, PPV of 86.21%, NPV of 97.82%, and F1-score of 85.87%. These findings demonstrate that integrating decisions from multiple classifiers significantly improves diagnostic accuracy and robustness.
