Office of Technology Transfer – University of Michigan

Automated Analysis of Optic Disk in Retinal Autofluorescence Images

Technology #6530

Questions about this technology? Ask a Technology Manager

Download Printable PDF

Categories
Researchers
Kanishka Thiran Jayasundera
Managed By
Drew Bennett
Associate Director - Software Licensing 734-615-4004
Patent Protection
US Patent Pending 2016-0345819
US Patent Pending

Fundus autofluorescence (FAF) imaging has proven a versatile method for non-invasive imaging of the retina. Information gained from this method is able to diagnose and monitor multiple diseases of the eye, including age-related macular degeneration (AMD), which affects over 5 million individuals worldwide. Retinal diseases are known to manifest during the later stages of life and the market for their treatment is expected to grow at a CAGR of over 11% as the worldwide population ages. To ensure market penetration of this imaging modality image processing methods are needed to adapt the analysis process to no longer require doctor intervention. Until now this has proven a formidable task due to the image-to-image variability and multiple features present in FAF images. The technology presented here utilizes a multi-step automated algorithm to segment FAF images into their key features. This eliminates the need for intervention by the doctor, and allows for easy quantification and monitoring of retinal disease.

Overview of Segmentation Algorithm

The automated segmentation of FAF images is achieved by first characterizing the main features in the image by applying a Gabor transformation to the illumination corrected image. These features are then clustered hierarchically utilizing their Euclidian distances to define groups. These clustered regions are fit separately to geometrical functions allowing the approximation of each feature’s boundary. The illumination corrected image is also processed to detect the direction, density, and length of the vessels in the field of view. This information, combined with the boundaries of the features, is processed to locate the optic nerve and extract it from the other features. Finally a level set algorithm utilizes the information on the identified optic nerve, vessels, and remaining features to trace each component in the image and quantify the results. This method has been shown to correctly identify the optic nerve in the presence of significant geographic atrophy and serves as a robust method for automated image processing of FAF images.

Applications

  • Automated fundus autofluorescence image processing
  • Quantification of geographic atrophy in age-related macular degeneration

Advantages

  • Fully automated method for FAF image analysis
  • Allows for telemedical approach toward AMD monitoring by FAF
  • Automated quantification of features removes human error from image interpretation
  • Algorithm easy discerns between optic nerve and atrophy relying on anatomy of the eye instead of threshold methods