Routine laboratory biomarker signature combined with machine learning for rapid differentiation of intracerebral hemorrhage from acute ischemic stroke
- 1Department of Clinical Laboratory Medicine, Xijing Hospital, Fourth Military Medical University
Abstract
Background: Rapid and reliable differentiation between intracerebral hemorrhage (ICH) and acute ischemic stroke (IS) is essential for appropriate triage and treatment. We aimed to develop and validate a laboratory-based machine learning model using routine clinical chemistry and hematology parameters to discriminate ICH from IS.
Methods: We retrospectively analyzed the clinical and laboratory data of 12,213 hospitalized stroke patients (including 3,251 with ICH and 8,962 with IS) and 2,893 hypertensive controls at Xijing Hospital between 2013 and 2023. A temporal validation cohort comprising 153 ICH and 342 IS patients admitted from January to December 2024 was also established. To address class imbalance, BorderlineSMOTE-2 (BS2) was applied. Candidate variables were screened using three feature selection strategies, including RFECV-ADA, Lasso, and Boruta. Six supervised machine learning models were trained using ten-fold cross-validation. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. Feature importance was interpreted using SHAP analysis. Mendelian randomization and colocalization analyses were performed as exploratory complementary analyses.
Results: Ten routine laboratory parameters were retained for model construction: white blood cell count, neutrophil percentage, cystatin C, uric acid, total protein, potassium, sodium, chloride, fibrinogen degradation product, and D-Dimer. In the test cohort, the final model (BS2_LightGBM_V10) achieved AUCs of 0.926 for discriminating ICH from IS, 0.979 for discriminating ICH from controls, and 0.923 for discriminating IS from controls. For the three-class classification task (ICH, IS, and controls), the overall accuracy reached 79.07%. SHAP analysis indicated that neutrophil percentage, cystatin C, D-Dimer, and total protein were the most influential predictors. Mendelian randomization analysis suggested that uric acid had a potential causal association with the risk of ICH (OR = 1.309), whereas uric acid (OR = 1.0008), cystatin C (OR = 1.0011), and potassium (OR = 1.0040) had potential causal associations with the risk of IS. Colocalization analysis identified 18 genes linked to uric acid in the context of ICH.
Conclusion: A diagnostic model based on ten routine laboratory indicators accurately differentiated ICH from IS. Uric acid may also play a causal role in the pathogenesis of both disorders.
Related
- Cross-trait genetic architecture of intracerebral hemorrhage highlights convergent vascular, hematologic and lipid pathways
- PINK1 agonist MTK458 attenuates cerebral edema by protecting blood-brain barrier integrity in intracerebral hemorrhage mice
- An automated, explainable, NCCT-based clinical decision-support system for spontaneous intracerebral hemorrhage.
- Development and Validation of a Human Cadaveric Model for Lobar Intracerebral Hemorrhage
- Admission White Blood Cell Count and Neutrophil‐to‐Lymphocyte Ratio for Early Risk Stratification after Spontaneous Intracerebral Hemorrhage: A Retrospective Cohort Study
BibTeX
@article{hao2026routine,
title = {{Routine laboratory biomarker signature combined with machine learning for rapid differentiation of intracerebral hemorrhage from acute ischemic stroke}},
author = {Xiaoyan Hao and Juan Wang and Jing Wang and Yujiao Hu and Fengjuan Wang and Xinran Liu and Congxia Bai},
journal = {Frontiers in Neurology},
year = {2026},
month = oct,
doi = {10.3389/fneur.2026.1931764},
url = {https://www.frontiersin.org/articles/10.3389/fneur.2026.1931764/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Routine laboratory biomarker signature combined with machine learning for rapid differentiation of intracerebral hemorrhage from acute ischemic stroke
AU - Xiaoyan Hao
AU - Juan Wang
AU - Jing Wang
AU - Yujiao Hu
AU - Fengjuan Wang
AU - Xinran Liu
AU - Congxia Bai
PY - 2026
DA - 2026/10/15
JO - Frontiers in Neurology
DO - 10.3389/fneur.2026.1931764
UR - https://www.frontiersin.org/articles/10.3389/fneur.2026.1931764/full
AB - Background: Rapid and reliable differentiation between intracerebral hemorrhage (ICH) and acute ischemic stroke (IS) is essential for appropriate triage and treatment. We aimed to develop and validate a laboratory-based machine learning model using routine clinical chemistry and hematology parameters to discriminate ICH from IS. Methods: We retrospectively analyzed the clinical and laboratory data of 12,213 hospitalized stroke patients (including 3,251 with ICH and 8,962 with IS) and 2,893 hypertensive controls at Xijing Hospital between 2013 and 2023. A temporal validation cohort comprising 153 ICH and 342 IS patients admitted from January to December 2024 was also established. To address class imbalance, BorderlineSMOTE-2 (BS2) was applied. Candidate variables were screened using three feature selection strategies, including RFECV-ADA, Lasso, and Boruta. Six supervised machine learning models were trained using ten-fold cross-validation. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. Feature importance was interpreted using SHAP analysis. Mendelian randomization and colocalization analyses were performed as exploratory complementary analyses. Results: Ten routine laboratory parameters were retained for model construction: white blood cell count, neutrophil percentage, cystatin C, uric acid, total protein, potassium, sodium, chloride, fibrinogen degradation product, and D-Dimer. In the test cohort, the final model (BS2_LightGBM_V10) achieved AUCs of 0.926 for discriminating ICH from IS, 0.979 for discriminating ICH from controls, and 0.923 for discriminating IS from controls. For the three-class classification task (ICH, IS, and controls), the overall accuracy reached 79.07%. SHAP analysis indicated that neutrophil percentage, cystatin C, D-Dimer, and total protein were the most influential predictors. Mendelian randomization analysis suggested that uric acid had a potential causal association with the risk of ICH (OR = 1.309), whereas uric acid (OR = 1.0008), cystatin C (OR = 1.0011), and potassium (OR = 1.0040) had potential causal associations with the risk of IS. Colocalization analysis identified 18 genes linked to uric acid in the context of ICH. Conclusion: A diagnostic model based on ten routine laboratory indicators accurately differentiated ICH from IS. Uric acid may also play a causal role in the pathogenesis of both disorders.
SN - 1664-2295
ER -