Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
Intracerebral Hemorrhage Hematoma Expansion Prediction with Deep Learning
18 Pages Posted: 29 Jul 2021
More...Abstract
Background: Hematoma expansion (HE) is associated with poor clinical outcomes following intracerebral hemorrhage (IcH). Existing predictors of HE from non-contrast computed tomography (NCCT) have low sensitivity and high inter-observer variability. Availability of an objective, standardized method to predict HE remains an unmet need.
Methods: This multicenter observational cohort study analyzed a retrospective dataset of 2350 patients (December 2011 to June 2018, NCCT only, 86 centers), and a prospective dataset of 460 patients (March 2018 to February 2019, NCCT plus clinical data, 28 centers) with acute IcH . Prediction of HE, defined as a hematoma growth of > 6ml within 48 hours was the primary outcome of the study. A deep learning system (DLS) was developed and validated to autonomously predict HE based on the NCCT images using retrospectively collected data. A multivariate logistic regression model and a five-point score based on clinical variables and DLS prediction score were then developed using prospectively collected data.
Findings: The DLS, using only NCCT, achieved HE prediction AUC of 0·783 (95% CI, 0·693-0·871) on the prospective validation dataset. Multivariate analysis of clinical variables indicated a higher risk of HE in patients with higher baseline NIHSS (OR: 4·37(1·56-12·25)), shorter onset-to-NCCT time (OR: 2·92(1·13-7·56)), and lack of antihypertensive therapy history (OR: 2·61 (1·05-6·53)) . Inclusion of these clinical variables along with DLS prediction, further improved the HE prediction accuracy, achieving an AUC of 0·812 (0·732-0·891) which was higher than all the previous methods. Finally, an easy-to-use five-point score was developed, which achieved an AUC of 0·789 (0·707-0·871).
Interpretation: Deep learning systems could provide an automated, objective, multicenter generalizable, and observer bias-free solution for HE prediction from NCCT images. Inclusion of clinical factors may further improve the prediction accuracy and the system can be used to identify patients at risk of HE leading to a personalized treatment planning.
Funding: Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences, Beijing Municipal Committee of Science and Technology, Ministry of Science and Technology of the People’s Republic of China, National Natural Science Foundation of China, and Beijing Nova Program.
Declaration of Interest: None to declare.
Ethical Approval: This study was approved by the Institutional Review Board of Beijing TianTan hospital which covers all the hospitals contributing data to this study as part of the Chinese Stroke Center Alliance
Suggested Citation: Suggested Citation