The Impact of Using BI-RADS with Voting Classifier Fusion for Early Detection of Breast Cancer

Breast cancer is one of the leading causes of death among women worldwide. The chances of a patient’s cure depend significantly on the stage of cancer at the time of diagnosis. Currently, the accuracy of Computer Aided Diagnosis System (CADs) using Mammography BI-RADS is approximately 80%, highlighting the need for further enhancement. The most commonly used classifiers include Multilayer Feed Forward Neural Network (MFFNN), Liner Discriminant Analysis (LDA), Support Vector Machine (SVM) and k-Nearest Neighbour (KNN). One of the key factors affecting the performance of CADs is the choice of classifier, as different classifiers may produce varying of misclassified samples, leading to differences in classification accuracy. In this study, 191 samples were used as an input for four classifiers: MFFNN, LDA with leave one out cross validation (LOOCV), SVM-LOOCV and KNN-LOOCV. To minimize the number of missclassified samples, this study employed the majority voting classifier fusion algorithm, which combines the results of the four classifiers to obtain a more robust decision. By comparing the accuracy of the majority voting classifier fusion algorithm with the best results obtained by individual classifier (KNN), the results showed that the voting classifier fusion algorithm outperformed KNN, achieving classification accuracies of 84.8% and 83.7%, respectively.