This approach stands out by integrating both 2D fundus and 3D depth information from OCT data, rather than relying solely on en face projections or retinal layer segmentation. The combined multi-scale features and adaptive boosting classifier deliver superior sensitivity to vessel structures under varying image quality and pathology. Matched filtering and optic nerve head masking further enhance specificity by reducing noise and anatomical artifacts. As a result, it achieves robust, high-resolution vessel segmentation well suited for clinical analysis, automated registration and early disease detection.
Description
The subject invention is an automated retinal blood vessel segmentation technique designed and developed based on 3D SD-OCT, with the aim to provide accurate vessel patterns for clinical analysis, retinal image registration, early diagnosis and monitoring of the progression of glaucoma and other retinal diseases. The technique uses machine learning algorithms to automatically identify blood vessels on 3D OCT images and does not rely on any other processing. SD-OCT is a new high resolution imaging technique, capable of achieving micrometer resolution in depth. It allows detailed imaging of the eye structures. Currently there is no independent method of retinal blood vessel segmentation on SD-OCT. The present invention provides an automated method to identify blood vessels on 3D OCT image without relying on any processing such as retinal layer segmentation.
Applications
Automated diabetic retinopathy screening
Glaucoma progression monitoring
Ophthalmic imaging software licensing
Teleophthalmology vessel analysis
Clinical trial imaging biomarker
Advantages
Automated 3D vessel segmentation reducing manual effort and variability
Improved accuracy and robustness by combining 2D and 3D features
Enhanced vessel visibility in OCT fundus images for clearer diagnostics
Robust performance across varied image qualities and retinal pathologies
Supports early detection and monitoring of diseases like glaucoma and diabetic retinopathy
Eliminates the need for explicit retinal layer segmentation
Speeds up clinical workflow and facilitates reliable image registration
IP Status
https://patents.google.com/patent/US8831304B2
