Abstract
BACKGROUND: Due to the high morbidity and mortality of primary intracerebral hemorrhage (ICH), several non-contrast computed tomography (NCCT) imaging markers were proposed to determine the prognosis of affected patients. We prospectively evaluated the predictive accuracy of certain imaging features and established a predictive model composed of highly relevant imaging and clinical features to identify the 3-month functional outcome in primary ICH patients. METHODS: Patients admitted for primary ICH to a tertiary care center (Al-Zahra Hospital, Isfahan, Iran) were prospectively included from September 2021 to October 2023. Inclusion criteria were defined as: Patients aged ≥18 years with primary or spontaneous ICH confirmed on NCCT at the time of admission. The baseline NCCT was conducted in the early stage of ICH (within 6 hours from symptom onset). The initial NCCT images were obtained within 6 hours from symptom onset. After 3 months, functional outcome of patients was assessed using the modified Rankin Scale (mRS); with mRS ≥3 as poor prognosis and mRS ≤2 as favorable prognosis. The Chi-squared and Logistic regression tests were used for determining the association between clinical and imaging features in differentiating patients' prognosis. Machine learning algorithm [support vector machine (SVM)] was also used to determine the importance rate of each relevant imaging sign in predicting prognosis. RESULTS: A total of 203 primary ICH patients were included, among which 119 patients (58.6%) had unfavorable prognosis at 3 months. Age, diastolic blood pressure, and Glasgow Coma Scale (GCS) score at admission were significantly associated with prognosis. Among imaging features, hemorrhage volume [95% confidence interval (CI): 0.972-0.991, P<0.001], the presence of midline shift (95% CI: 2.038-7.911, P<0.001), blend sign (95% CI: 1.081-3.760, P=0.026), satellite sign (95% CI: 1.451-4.764, P=0.001), and black hole sign (95% CI: 2.262-12.714, P<0.001) were significantly different among 2 groups. SVM algorithm showed hemorrhage volume the most important prognostic imaging feature (importance rate: 100%), along with black hole (63.1%), midline shift (54%), satellite (20.4%), and blend sign (15.6%); with decreasing order of importance. CONCLUSIONS: Using certain radiological and clinical features, we established a model with considerable prognostication in management of patients with primary ICH in emergency departments.