Machine Learning Approaches to Radiogenomics of Breast Cancer using Low-Dose Perfusion Computed Tomography: Predicting Prognostic Biomarkers and Molecular Subtypes

Eun Kyung Park,Gil Soo Son,Hye Yoon Lee,Bo Kyoung Seo,Ok Hee Woo,Kyu Ran Cho,Kwang-Sig Lee,Young Woo Chang

doi:10.1038/s41598-019-54371-z

Eun Kyung Park, Gil Soo Son + Show 6 more

Open Access

https://doi.org/10.1038/s41598-019-54371-z

Copy DOI

Abstract

Radiogenomics investigates the relationship between imaging phenotypes and genetic expression. Breast cancer is a heterogeneous disease that manifests complex genetic changes and various prognosis and treatment response. We investigate the value of machine learning approaches to radiogenomics using low-dose perfusion computed tomography (CT) to predict prognostic biomarkers and molecular subtypes of invasive breast cancer. This prospective study enrolled a total of 723 cases involving 241 patients with invasive breast cancer. The 18 CT parameters of cancers were analyzed using 5 machine learning models to predict lymph node status, tumor grade, tumor size, hormone receptors, HER2, Ki67, and the molecular subtypes. The random forest model was the best model in terms of accuracy and the area under the receiver-operating characteristic curve (AUC). On average, the random forest model had 13% higher accuracy and 0.17 higher AUC than the logistic regression. The most important CT parameters in the random forest model for prediction were peak enhancement intensity (Hounsfield units), time to peak (seconds), blood volume permeability (mL/100 g), and perfusion of tumor (mL/min per 100 mL). Machine learning approaches to radiogenomics using low-dose perfusion breast CT is a useful noninvasive tool for predicting prognostic biomarkers and molecular subtypes of invasive breast cancer.

Highlights

The use of computed tomography (CT) to image the breast is limited because of the risk of high radiation exposure and poor image quality
In terms of accuracy and the area under the receiver-operating characteristic curve (AUC), the random forest model was the best model for predicting prognostic biomarkers and molecular subtypes of breast cancer
The accuracy was higher for the random forest model than for the logistic regression model by 13% on average: 78% vs. 62% for lymph node status, 80% vs. 67% for tumor grade, 77% vs. 64% for tumor size, 82% vs. 70% for estrogen receptor (ER) status, 78% vs. 66% for progesterone receptor (PR) status, 83% vs. 78% for human epidermal growth factor receptor 2 (HER2) status, 77% vs. 65% for Ki67 status, and 66% vs. 48% for molecular subtypes

Summary

Introduction

The use of computed tomography (CT) to image the breast is limited because of the risk of high radiation exposure and poor image quality. Low-dose breast perfusion CT requires a short scan time of only 3 minutes, produces satisfactory image quality, and can, be applied to patients who cannot undergo MRI because of implantable metallic devices, allergy to gadolinium, marked obesity, severe kyphoscoliosis, or claustrophobia. A few studies have applied a machine learning approach to radiogenomics and predictive analysis[10,11]. The purpose of our study was to investigate the clinical value of a machine learning approach to radiogenomics using low-dose perfusion breast CT for predicting prognostic biomarkers and molecular subtypes in invasive breast cancer. The concentration of contrast agent within an organ at any time will have intravascular and extravascular components. The intravascular component is determined by the blood volume of the organ and the blood concentration of the contrast agent used at that time. We tried to identify the best machine learning model and which CT parameters are important for predicting prognostic biomarkers and molecular subtypes

Objectives

Methods

Results

Discussion

Conclusion