AUTHOR=Gao Min , Huang Siying , Pan Xuequn , Liao Xuan , Yang Ru , Liu Jun 

TITLE=Machine Learning-Based Radiomics Predicting Tumor Grades and Expression of Multiple Pathologic Biomarkers in Gliomas

JOURNAL=Frontiers in Oncology

VOLUME=Volume 10 - 2020

YEAR=2020

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2020.01676

DOI=10.3389/fonc.2020.01676

ISSN=2234-943X

ABSTRACT=Abstract
Background The grading and pathologic biomarkers of glioma are of guiding significance for individual treatment of glioma. Clinically, it is often necessary to obtain pathological samples through invasive operation. We aim to use a variety of machine learning algorithms to automatically predict tumor grades and pathologic biomarkers through non-invasive traditional MRI detection methods.
Methods We retrospectively collected a data set of 367 glioma patients with pathological reports and had MRI scans between October 2013 and March 2019. Using enhanced MRI images as input data, we built three frequently-used machine-learning based models, LC, SVM, and RF, for four predictive tasks: 1) glioma grades, 2) Ki67 expression level, 3) GFAP expression level, and 4) S100 expression level in gliomas. Each sub dataset was divided into training and testing sets by the radio of 4 to 1.
Results Prediction of histological grade of glioma, the RF classifier achieved a good predictive performance (AUC: 0.79, accuracy: 0.81). The RF classifier achieved good predictive performance on Ki67 expression measured in AUC (0.85), accuracy (0.80). The RF classifier achieved a fair predictive performance on GFAP expression measured in: AUC (0.72), accuracy (0.81). Specifically, for S100 low expression level: the accuracy (0.95), sensitivity (0.94), specificity (0.97), f1 (0.95); while for high expression level, none of 4 high expression cases was predicted correctly.
Conclusions A machine-learning based radiomics approach provides a non-invasive tool for the prediction of glioma grades and expression levels of multiple pathologic biomarkers preoperatively with favorable predictive accuracy and stability.