Detection of Stroke with Retinal Microvascular Density and Self-Supervised Learning Using OCT-A and Fundus Imaging

Area of Science:

• Ophthalmology
• Neurology
• Medical Imaging

Background:

• Acute cerebral stroke is a major cause of death and disability.
• Prompt diagnosis during transport can improve outcomes.
• Retinal imaging offers a potential non-invasive method for stroke detection due to shared vascular similarities.

Purpose of the Study:

• To investigate the feasibility of using retinal imaging for acute cerebral stroke diagnosis.
• To identify effective imaging features and modalities for stroke detection.
• To train machine learning models for classifying stroke patients using retinal data.

Main Methods:

• Utilized Optical Coherence Tomography Angiography (OCT-A) and fundus images.
• Extracted features using traditional methods (feature engineering) and self-supervised deep learning.
• Trained machine learning models to differentiate between control subjects and acute stroke patients.

Main Results:

• Models using macular microvasculature density features achieved an AUC of 0.87-0.88.
• Self-supervised learning models yielded AUCs ranging from 0.66 to 0.81.
• Initial evidence suggests OCT-A microvasculature density is a promising diagnostic indicator.

Conclusions:

• Retinal microvasculature density features show potential for diagnosing acute cerebral stroke.
• Further research is required to develop a definitive diagnostic system.
• This approach could enable faster stroke diagnosis during patient transportation.