A Novel Prediction Framework for Two-Year Stroke Recurrence Using Retinal Images

Stroke is a malignant disease with high incidence rate, high disability rate and high mortality rate. Particularly, the incidence rate of stroke recurrence is much higher than that of initial stroke. It brings a heavy life and financial burden to patients and their family. In this paper, we propose a novel prediction framework for two-year stroke recurrence based on features extracted from retinal images. Specifically, 425 patients with initial stroke were recruited from Shenzhen Traditional Chinese medicine Hospital and collected their clinical and retinal images between January 2017 and January 2019. After follow-up, 103 patients had stroke recurrence within 2 years. All collected retinal images are analyzed and the characteristics of fundus vessels are extracted by an automatic retinal image analysis system. We employ four widely used machine learning methods of support vector machine (SVM), random forest (RF), logistic regression (LR), and XGBoost to predict two-year recurrent stroke events. Experiment results show that our proposed prediction framework for two-year recurrent stroke can achieve promising results, the best prediction accuracy is up to 84.38%. Our proposed framework for predicting two-year stroke recurrence can be potentially applied in medical information systems to predict malignant events and help medical providers to take intervention in advance to prevent malignant events.