AUTHOR=Li Qingrun , Li Feng , Liu Hao , Li Yan , Chen Hongri , Yang Wenrui , Duan Shaofeng , Zhang Hongying 

TITLE=CT-based radiomics models predict spontaneous intracerebral hemorrhage expansion and are comparable with CT angiography spot sign

JOURNAL=Frontiers in Neurology

VOLUME=Volume 15 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2024.1332509

DOI=10.3389/fneur.2024.1332509

ISSN=1664-2295

ABSTRACT=Background and purpose: This study aimed to investigate the efficacy of radiomics, based on non-contrast computed tomography (NCCT) images and computed tomography angiography (CTA) images, in predicting early hematoma expansion (HE) in patients with spontaneous intracerebral hemorrhage (SICH). Additionally, the predictive performance was compared with the established CTA spot sign.: A retrospective analysis was conducted on CT images of 182 patients with SICH. The patients were divided into a training set (145 cases) and a testing set (37 cases) using random stratified sampling. Two radiomics models were constructed by combining quantitative features extracted from NCCT images (NCCT model) and CTA images (CTA model) using a logistic regression (LR) classifier. Additionally, a univariate LR model composed of the CTA spot sign (spot sign model) was established. The predictive performance of the two radiomics models and the spot sign model was compared using the area under the curve (AUC) of the receiver operating characteristic (ROC). Results: In the training set, the AUCs of the NCCT model, CTA model, and spot sign model were 0.938, 0.904, and 0.726, respectively. Both the NCCT model and the CTA model demonstrated superior predictive performance compared to the spot sign model (all P < 0.001), with the performance of the two radiomics models being comparable (P = 0.068). In the testing set, the AUCs were 0.925, 0.873, and 0.720, respectively, with only the NCCT model exhibiting a significantly greater capability than the spot sign model (P = 0.041). Conclusion: Radiomics models based on NCCT and CTA images effectively predicted HE in patients with SICH. The predictive performance of the NCCT and CTA models were similar, with the NCCT model outperforming the spot sign model. These findings suggested the potential to reduce the need for CTA examinations, thereby reducing radiation exposure and the usage of contrast agents in future hematoma expansion prediction practices.