Machine Learning Models for Histopathological Breast Cancer Image Classification

Breast cancer is among the most common forms of cancer, representing 30% of cancer cases in the United States per year and is the leading cause of cancer-induced mortality. Diagnosis is determined through biopsies followed by analysis of histopathological images. Machine Learning (ML) proves to be an essential image classification tool, with Convolutional Neural Networks (CNN) providing high accuracy, but long training time. To provide optimal ML models for medical use and evaluate model performance, unstructured images are converted to structured data. Singular Value Decomposition extracted features from 277,524 histopathological images and Recursive Feature Selection reduced the dimensionality of the resultant dataset. Logistic Regression, Decision Tree, XGBoost, K-Nearest neighbours and Multi-layer Perceptron ML models were trained and evaluated. K-Nearest neighbours had the highest accuracy of 77.94%. While these models do not have optimal accuracy for real world applications, their exceptional speed provides support for further improvement and exploration.