Breast Cancer Molecular Subtype Detection through Mammograms with Machine Learning: A Comprehensive Framework Using Radiomics and Metaheuristic Optimization

Abstract

Machine learning has become pivotal in devising effective solutions for intricate tasks such as cancer detection, owing to its capacity to process extensive datasets with remarkable efficiency and accuracy. This study emphasizes the importance of molecular subtype detection in breast cancer diagnosis. Using the CMMD dataset, we applied various image pre-processing techniques to refine detection capabilities such as cropping, filtering, enhancement, and normalization. The radiomics features such as gray level run length matrix, local binary pattern, and gray level co-occurrence matrix were extracted from the image and optimized using differential evolution (DE) and grey wolf optimization. The evaluation of support vector machine, decision tree, light gradient boosting machine (LGBM), and random forest classifiers revealed that LGBM consistently demonstrated superior performance in most classification tasks. Specifically, when using a median filter and optimized with DE: LGBM achieved an accuracy of 0.81 in predicting the benign or malignant status of breast cancer for coarse-grained classification task and among the fine-grained classification tasks when predicting calcifications that are either benign or malignant, LGBM obtained an accuracy of 0.83. These findings highlight the potential of machine learning, particularly the LGBM classifier, in enhancing breast cancer diagnosis.

Topics & Concepts