AUTHOR=Liu Wei , Jia Longbin , Xu Lina , Yang Fengbing , Guo Zixuan , Li Jinna , Zhang Dandan , Liu Yan , Xiang Han , Cheng Hongjiang , Hou Jing , Li Shifang , Li Huimin 

TITLE=Prediction of early neurologic deterioration in patients with perforating artery territory infarction using machine learning: a retrospective study

JOURNAL=Frontiers in Neurology

VOLUME=Volume 15 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2024.1368902

DOI=10.3389/fneur.2024.1368902

ISSN=1664-2295

ABSTRACT=Background: Early neurologic deterioration (END) is a frequently complication in pat ients with perforating artery territory infarction (PAI), leading to poorer outcomes. Th erefore, we aimed to apply machine learning (ML) algorithms to predict the occurrenc e of END in PAI and investigate related risk factors.Methods: This retrospective study analyzed a cohort of PAI patients, excluding those with severe stenosis of the parent artery. We included demographic characteristics, cli nical features, laboratory data, and imaging variables. Recursive feature elimination w ith cross-validation (RFECV) was performed to identify critical features. Seven ML al gorithms (logistic regression, random forest, adaptive boosting, gradient boosting deci sion tree, histogram-based gradient boosting, extreme gradient boosting, category boo sting) were developed to predict END in PAI patients using these critical features. We compared the accuracies of these models in predicting outcomes. Additionally, SHapl ey Additive exPlanations (SHAP) values were introduced to interpret the optimal mod el and assess the significance of input features.The study enrolled 1020 PAI patients with a mean age of 60.46 (range 49.11-71.81) years. Of these, 30.39% were female, and 129 (12.65%) experienced END. RF ECV selected thirteen critical features, including blood urea nitrogen (BUN), total cho lesterol (TC), low-density-lipoprotein cholesterol (LDL-C), apolipoprotein B (apoB), atrial fibrillation, loading dual antiplatelet therapy (DAPT), single antiplatelet therapy (SAPT), argatroban, basal ganglia, thalamus, posterior choroidal arteries, maximal axi al infarct diameter (measured at <15mm) and stroke subtype. The gradient boosting de cision tree had the highest area under the curve (0.914) among the seven ML algorith ms. The SHAP analysis identified apoB as the most significant variable for END.Our results suggest that ML algorithms, especially the gradient boosting decision tree, are effective in predicting the occurrence of END in PAI patients.