Machine learning versus deep learning for screening ischemic stroke among asymptomatic population.

Ischemic stroke puts great health burden in public. However, the diagnosis is based on head CT or MRI scanning. We aim to develop classifier models to screen asymptomatic ischemic stroke based on clinical data without imaging. The subjects (age ≥ 20 years) were recruited into our study who attended health examination. We developed models to classify subjects with or without ischemic stroke using support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), and convolutional neural network (CNN). The candidate variables included sex, age, height, weight, blood pressure, blood lipid levels, fasting plasma glucose, hyperlipidemia, smoke history, alcohol drinking, diabetes, hypertension and coronary heart disease. Models included 141 subjects with small artery occlusion (SAO), 70 with large-artery atherosclerosis (LAA) and 211 controls. Classifier models to diagnose ischemic stroke, and further SAO and LAA, were developed with machine learning and deep learning algorithms. The best performance of model was based on RF, followed by CNN. The blood pressure, blood lipid and fasting plasma glucose were important factors and contributed to classifier models with machine learning algorithms. In conclusion, classifier models for ischemic stroke could be developed with machine learning and deep learning algorithms based on laboratory testing data without imaging. The models with RF showed best performance, followed by CNN. Those models may provide a tool for primary care doctors to screen asymptomatic stroke in general population, which could benefit to stroke patients from early diagnosis.