DBL-MPE: Deep Broad Learning for Prediction of Response to Neo-adjuvant Chemotherapy Using MRI-Based Multi-angle Maximal Enhancement Projection in Breast Cancer

Abstract

Neo-adjuvant chemotherapy (NAC) is one of the main treatments in breast cancer, given before surgery to reduce the tumor's size and increase the surgical outcome's success rate. Predicting the response of breast cancer patients to NAC can be challenging, and inaccurate prediction may lead to suboptimal treatment outcomes. Previous studies have shown that machine learning methods based on Magnetic Resonance Imaging (MRI) can be used to predict the response of breast cancer to NAC with promising accuracy. However, data heterogeneity and feature representation dilemmas are still significant challenges. In this paper, we propose a novel framework Deep Broad Learning Maximal Enhancement Projection (DBL-MEP) with two main functional modules, that is, maximal enhancement projection (MEP) maps and deep-broad learning (DBL). Firstly, the framework can transform dynamic contrast enhancement MRI (DCE-MRI) to acquire multi-angle MEP maps, which can reduce the effect of heterogeneity caused by data collected from different centers. Secondly, the framework uses DBL for feature extraction and integration, which can reduce the risk of overfitting by using a diverse set of imaging features. The framework is trained and tested on a large-scale dataset with 1589 breast cancer patients from multiple centers. Extensive experiments demonstrate that the proposed method is superior to traditional method and shows stable performance across different scanners and field strengths. To the best of the authors’ knowledge, this is the first paper to apply deep broad learning techniques and multi-angle enhancement projection maps of DCE-MRI in prediction of treatment response to NAC in breast cancer.