Stroke subtypes risk prediction and detection using retinal vascular structure and oxygen saturation analysis.

BACKGROUND

Stroke presents a substantial health burden, emphasizing the crucial necessity for robust screening tools to gauge its severity and assess associated biomarkers. The identification of retinal biomarkers for stroke is a pivotal pursuit, enabling both early-stage risk and subsequent prognosis prediction for personalized intervention strategies. This study aims to analyze retinal blood vessel oxygen saturation (SO) and structure across ischemic and hemorrhagic stroke subtypes compared to healthy controls. Additionally, it seeks to examine adjusted odds ratios between retinal vascular features and two stroke subtypes, employing these metrics for the classification of stroke.

METHODS

This study assessed retinal images from 29 ischemic stroke patients, 23 hemorrhagic stroke patients, and 82 controls. SO levels in both arteries and veins were assessed across all groups, marking a pioneering exploration into the distinctive subtypes of stroke. Using statistical and deep learning techniques, we also uniquely performed comprehensive structural vascular analysis in hemorrhagic stroke patients. Logistic regression identified relationships between retinal biomarkers and stroke types. Random forest classification differentiated stroke and control based on these retinal vascular biomarkers.

RESULTS

Ischemic stroke patients exhibited significantly higher arterial SO compared to controls (P<0.01), while hemorrhagic patients showed no differences (P=0.34). Both stroke groups had reduced arterial density (ischemic . controls: P<0.01; hemorrhagic . controls: P<0.01) and fractal dimensions (ischemic . controls: P<0.01; hemorrhagic . controls: P<0.01). The results of logistic regression analysis indicated a discernible relationship between these biomarkers and the occurrence of both types of strokes. Integrating functional SO and structural biomarkers enabled over 80% accurate classification of stroke from retinal images.

CONCLUSIONS

Our study reveals marked differences in retinal blood vessel characteristics between stroke subtypes and controls. Through logistic regression analysis, we establish a robust association between these parameters and the incidence of both ischemic and hemorrhagic strokes, enhancing our ability to anticipate stroke risk. Subsequently, we showcase the prognostic potential of retinal vascular biomarkers by innovatively analyzing retinal images through machine learning for stroke occurrence. These findings suggest that retinal biomarkers may hold potential value for risk stratification in stroke, and with further investigation, could inform broader applications in cerebrovascular health.