Enhancing the Accuracy of Breast Cancer Detection with a Hybrid Clustering Algorithm Combining K-Means and GMM

Abstract

Abstract: Breast cancer is among the most widespread ailments afflicting women globally. The timely identification and accurate diagnosis are vital for successful therapy and improved prognosis. Over the past years, researchers have extensively utilized machine learning algorithms to detect breast cancer from medical images. This paper proposes an innovative hybrid clustering approach combining both k-means and Gaussian mixture models (GMM) to enhance breast cancer detection performance. By utilizing the k-means clustering approach as a basis, our algorithm generates initial cluster centers from the input data. With these results, we then proceed to implement the GMM algorithm to further refine our clustering outcomes and calculate each cluster's probability distribution accordingly. Through evaluation on publicly available mammography images, our hybrid algorithm outperformed both k-means and GMM algorithms in terms of sensitivity, specificity and area under the receiver operating characteristic curve (AUC-ROC).