Machine learning techniques for breast cancer diagnosis and treatment: a narrative review

Abstract

Objective: This narrative review describes the recent developments and applications of machine learning (ML), a part of artificial intelligence, concerning breast cancer. Background: The advent of new bioinformatic approaches and artificial intelligence-based computational technologies has led to a shift in the decision-making of oncologists regarding breast cancer diagnostics and treatment processes. Various successful applications of ML on image processing, especially the use of deep neural networks and convolutional neural networks, to detect tumor and lymph nodes regions have been reported. Recent high-throughput molecular quantifications, i.e., quantitative omics techniques have enabled simultaneous monitoring of thousands of molecules to understand the molecular-level pathology. These data, including gene expression, protein, metabolite, and methylation profiling, have been analyzed via deep learning, network analysis, clustering, and dimension reductions to explore intrinsic subtypes and new biomarkers. Clinical-pathological features have been conducted by multivariable analysis to predict various outcomes, e.g., the sensitivity of adjuvant therapy and prognosis. The quantitative relationships among their variables have been visualized as nomograms. To analyze complex structures of a larger number of variables, ML combining multiple clinical-pathological features has been developed to predict the prognosis, metastasis, and treatment outcomes of breast cancer. Methods: We provided the narrative review of ML-related topics especially in the quantitative omics data and clinical-pathological prediction models. Conclusion: ML-based prediction methods are powerful tools and contribute to realizing personalized medicine for breast cancer.