Radiomics-based prediction of intraoperative blood loss during intracranial meningiomas surgery.

Abstract

Meningiomas surgery is frequently accompanied by substantial blood loss, which is associated with an increased incidence of medical morbidities. Neuroimaging features, such as radiomic characteristics, could provide additional quantitative information on the tumor. Nonetheless, the usefulness of radiomics in predicting intraoperative blood loss has yet to be validated. Our objective was to examine the potential of radiomics to predict intraoperative blood loss in patients with intracranial meningiomas. A total of 137 patients with primary diagnosed intracranial meningiomas were evaluated via high-resolution brain magnetic resonance imaging (MRI), which included T1-weighted pre- and postcontrast imaging, T2-weighted imaging, diffusion-weighted (with apparent diffusion coefficient) imaging, and arterial spin labeling (ASL). MRI data were processed with subsequent extraction of radiomic features. The most significant predictors were determined via random forest regression analysis to model the relationships between selected metrics and the rate of intraoperative bleeding. We created a regression model based on ten radiomic predictors, including first- and second-order radiomic features. The resulting model allowed us to predict intraoperative blood loss in patients with intracranial meningiomas with a mean absolute error of 135.14 ml and R-squared value of 0.29, which could be considered good prediction quality. Tumor volume, tumor location, histological grade, and surgery duration were found to be less significant predictors than the other parameters and did not improve the model. Radiomic features could be useful in predicting intraoperative blood loss and provide valuable information for the presurgical evaluation of patients with intracranial meningiomas.