Advanced machine learning-based differentiation of acute ischemic and hemorrhagic stroke: ML-based stroke classification
Neurophysiology, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s11062-026-10035-z
- Dergi Adı: Neurophysiology
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, BIOSIS, EMBASE, Academic Search Ultimate (EBSCO), Natural Science Collection (ProQuest), Biological Science Database (ProQuest), Biomedical Reference Collection: Corporate Edition (EBSCO), Health Research Premium Collection (ProQuest), Pharma Collection (ProQuest)
- Anahtar Kelimeler: Biomarker Discovery, Clinical Diagnostics, Explainable Artificial Intelligence, Machine Learning, Stroke
- Atatürk Üniversitesi Adresli: Evet
Özet
Stroke, a cerebrovascular disease characterized by high mortality, escalating incidence, and mounting financial strain on healthcare systems, poses a significant global health challenge. While traditional regression-based analyses have been prevalent, the advent of machine learning (ML) techniques has unveiled a promising avenue for enhanced predictive capabilities, particularly in the realm of multi-parameter clinical blood indicators. 1,000 patients diagnosed with acute ischemic stroke were screened, and a total of 609 patients, including 330 with hemorrhagic and 279 with ischemic strokes, were retrospectively analyzed. Advanced machine learning methods like Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Adaptive Boosting (AdaBoost), Light Gradient Boosting Machine (LightGBM), and Hist Gradient Boosting Classifier (HGBCM) were employed to create predictive models for differentiating between ischemic and hemorrhagic strokes. The significance of blood biomarkers was assessed using explainable Artificial Intelligence (XAI) to interpret the prediction models with the maximum accuracy. The significance of blood biomarkers such as EOSP, sodium, GGT, glucose, LDH, creatine kinase, alkaline phosphatase, NEU, calcium, tGFH, and total protein in distinguishing between ischemic and hemorrhagic stroke was analyzed using Shapley values to assess the impact of each feature on the model’s prediction outcomes. Our findings illuminate EOSP, LDH, NEUP, tGFH, MONOP, RBC, ALT and BASOP as the paramount predictive values in both hemorrhagic and ischemic stroke subtypes. Furthermore, SHAP analysis indicated that lower EOSP values were associated with a higher predicted probability of both ischemic and hemorrhagic stroke in the developed models. Similarly, higher tGFH values contributed to an increased predicted probability of stroke subtype classification