Classification of Breast DCE-MRI Images via Boosting and Deep Learning Based Stacking Ensemble Approach

Abstract

The radiomics features are capable of uncovering disease characteristics to provide the right treatment at the right time where the disease is imaged. This is a crucial point for diagnosing breast cancer. Even though deep learning methods, especially, convolutional neural networks (CNNs) have demonstrated better performance in image classification compared to feature-based methods and show promising performance in medical imaging, but hybrid approaches such as ensemble models might increase the rate of correct diagnosis. Herein, an ensemble model, based on both deep learning and gradient boosting, was employed to diagnose breast cancer tumors using MRI images. The model uses handcrafted radiomic features obtained from pixel information breast MRI images. Before training the model these radiomics features applied to factor analysis to optimize the feature set. The accuracy of the model is 94.87% and the AUC value 0.9728. The recall of the model is 1.0 whereas precision is 0.9130. F1-score is 0.9545.