Radiomics-based left ventricular ejection fraction prediction: a feasibility study

Abstract

Objective·To assess the feasibility of using 3D imaging features extracted from cardiac magnetic resonance (CMR) short-axis cine images to predict left ventricular ejection fraction (LVEF).Methods·A total of 100 left ventricular hypertrophy (LVH) patients who visited the Department of Cardiology, Renji Hospital, Shanghai Jiao Tong University School of Medicine from January 2018 to December 2021, as well as 100 healthy control (HC) subjects during the same period, were included. All subjects completed CMR examinations under the supervision of experienced cardiologists and radiologists. The endocardial and epicardial contours were then manually delineated by cardiologists. Measurements of cardiac function and morphology were completed and data was recorded, including LVEF, left ventricular end-diastolic volume (LVEDV), and left ventricular end-diastolic mass (LVEDM). Myocardial 3D radiomic features of CMR-cine sequences were extracted by the Pyradiomics package, and selected and sorted by using correlation coefficient and K-best method. The LVEF prediction was performed with linear regression (LR), random forest (RF) and gradient boost (GB) methods. Results were also compared with LVEF prediction based on clinical information and CMR parameters.Results·In terms of clinical indicators, there were significant differences between the LVH and HC groups, such as LVEDV and LVEDM (all P<0.05); after extracting 3D radiomic features, the top 10 features were selected for further analysis. LR regression model, GB regression model and RF regression model were constructed for predicting the LVEF, and RF regression models showed the best results with seven features, in which the mean absolute error (MAE) was 0.066±0.002. Further comparison results showed that the model using radiomic information with CMR parameters (MAE=0.056±0.001) had the best performance and it was significantly better than the model using radiomic features (MAE=0.066±0.002) or CMR parameters (MAE=0.060±0.001) alone (both P<0.05).Conclusion·The use of radiomic features for LVEF prediction has certain feasibility, and combining radiomic features with CMR parameters can further improve the prediction accuracy of the model.

Keywords

• cardiac magnetic resonance (cmr)
• radiomics
• left ventricular hypertrophy
• cine sequence