Method for Confidence Metric in Optic Disk Location in Retinal Images
To improve accuracy in diagnosis of retinal disease, ORNL researchers invented a
method for assigning a confidence metric to computer-aided optic disc analysis. The
physical condition of the optic disk determines the presence of various ophthalmic
pathologies, including glaucoma and diabetic retinopathy. Unfortunately, localization
of the optic disk and detection of its boundaries on the retinal image are not easy tasks.
With this invention, the review process can be entirely automated.
Current methods for evaluation of retinal health require a patient to visit a highly
trained specialist or a retinal reading center that uses computer-aided assessment.
To improve this process, research is under way to develop automated techniques
to analyze retinal images; however, even the best methods misdiagnose diseases
between 5% and 20% of the time.
This invention offers two different methods to compare locations of the optic disk.
When the distance between two locations is above a threshold, the detection
is considered low confidence; when the distance is small, it is considered high
confidence. Depending on the conclusion, the method can flag the image for further
review by an ophthalmic professional.
- Potentially higher success rate of accurate optic disk detection on retinal images
- Rated confidence levels assignments
- Optic disk detection
- Disease screening
Thomas P. Karnowski, Kenneth W. Tobin, Jr., and Vijaya P. Muthusamy Govindasamy, Method for Confidence Metric in Optic Disk Location in Retinal Images, U.S Patent Application 12/263,876, filed November 3, 2008.
Thomas P. Karnowski
Measurement Science and Systems Engineering Division
Oak Ridge National Laboratory
|Title and Abstract||
METHOD AND SYSTEM FOR ASSIGNING A CONFIDENCE METRIC FOR AUTOMATED DETERMINATION OF OPTIC DISC LOCATION
A method for assigning a confidence metric for automated determination of optic disc location that includes analyzing a retinal image and determining at least two sets of coordinates locating an optic disc in the retinal image. The sets of coordinates can be determined using first and second image analysis techniques that are different from one another. An accuracy parameter can be calculated and compared to a primary risk cut-off value. A high confidence level can be assigned to the retinal image if the accuracy parameter is less than the primary risk cut-off value and a low confidence level can be assigned to the retinal image if the accuracy parameter is greater than the primary risk cut-off value. The primary risk cut-off value being selected to represent an acceptable risk of misdiagnosis of a disease having retinal manifestations by the automated technique.
|Oak Ridge National Laboratory||11/03/2008
|Technology ID||Development Stage||Availability||Published||Last Updated|