Prediction of Hematoma Expansion in Intracerebral Hemorrhage in 24 Hours by Machine Learning Algorithm.

OBJECTIVE: The significance of noncontrast computer tomography (CT) image markers in predicting hematoma expansion (HE) following intracerebral hemorrhage (ICH) within different time intervals in the initial 24 hours after onset may be uncertain. Hence, our objective was to examine the predictive value of clinical factors and CT image markers for HE within the initial 24 hours using machine learning algorithms. METHODS: Four machine learning algorithms, including extreme gradient boosting (XGBoost), support vector machine, random forest, and logistic regression, were employed to assess the predictive efficacy of HE within every 6-hour interval during the first 24 hours post -ICH. The area under the receiver operating characteristic curves was utilized to appraise predictive performance across various time periods within the initial 24 hours. RESULTS: A total of 604 patients were included, with 326 being male, and 112 experiencing hematoma expansion (HE). The findings from machine learning algorithms revealed that computed tomography (CT) image markers, baseline hematoma volume, and other factors could accurately predict HE. Among these algorithms, XGBoost demonstrated the most robust predictive model results. XGBoost's accuracy at different time intervals was 0.89, 0.82, 0.87, and 0.94, accompanied by F1 -scores of 0.89, 0.80, 0.87, and 0.93, respectively. The corresponding area under the curve was 0.96, affirming the precision of the predictive capability. CONCLUSIONS: Computed tomography (CT) imaging markers and clinical factors could effectively predict HE within the initial 24 hours across various time periods by machine learning algorithms. In the expansive landscape of big data and multimodal cerebral hemorrhage, machine learning held significant potential within the realm of neuroscience.