Advantages:
- A novel computerized system integrates entropy filtering and FFT-based scoring to enable automated, precise uveitis diagnosis
- A computerized method provides an objective, reproducible score for ocular inflammation, reducing subjectivity in clinical grading scales
- Utilize low pass filtering and image subtraction to highlight obscured fundoscopic details in diabetic retinopathy
- The use of an entropy filter enables a quantitative analysis of retinal vessel texture, providing detailed assessments for diabetic retinopathy
Summary:
In ophthalmology, assessing intraocular inflammation, such as vitreous haze, has traditionally relied on subjective photographic grading systems, which can lead to variability among clinical evaluators. This inconsistency poses a significant challenge, as accurate and reproducible measurements are essential for the proper diagnosis and monitoring of uveitis and similar conditions. Moreover, conventional imaging methods struggle to differentiate subtle details obscured by haze due to overlapping frequency patterns and the inherently complex texture of retinal structures. These factors motivate the need for more objective, automated approaches that quantitatively isolate and evaluate the critical spatial frequency components relevant to retinal blood vessel integrity while overcoming limitations in standard subjective assessments.
Our researcher introduces a novel automated method for grading vitreous haze by processing fundoscopic images through a series of steps that enhance obscured retinal features. Initially, a grayscale conversion and low-pass filtering remove high-frequency details and isolate the background, and subtracting this blurred image from the original produces a difference image that emphasizes hidden structures. Applying a sliding-window entropy filter, tuned to capture the characteristic size range of retinal blood vessels, quantifies local texture variations linked to retina clarity. Subsequently, a Fast Fourier Transform (FFT) of the entropy image yields a comprehensive power spectrum that is spatially averaged across multiple orientations and condensed into a one-dimensional representation. By integrating power within a specific frequency band corresponding to the spatial frequencies of retinal vessels, a blur score is generated and scaled to match clinical vitreous haze grading. This process provides a more precise method as it objectively correlates quantitative frequency-domain metrics with clinical assessments of intraocular inflammation, offering enhanced accuracy, reproducibility, and faster evaluation.
Set of fundus and entropy images with graphical comparisons of computed blur scores and expert gradings A) A set of four fundoscopic images from the trial set and four entropy images corresponding to the fundoscopic images. B) A graphical comparison of the present invention's computed blur scores with vitreous haze scores of a second expert grader. C) A graphical comparison of the present invention's computed blur scores with vitreous haze scores of a third expert grader.
Desired Partnerships: