Robust estimation of corneal limbus from the placido-disk videokeratoscopic images

Purpose Accurate knowledge of the corneal limbus is important when calculating corneal surface area, finding the corneal centroid and comparing it to the pupil centre or corneal apex. Existing techniques for estimating corneal limbus from digital photographs are based on manual measurements or computer based image processing. We aim to develope a robust and accurate technique that can be applied to videokeratoscope images routinely acquired during corneal topography measurements. Methods We have refined our recently reported soft edge detection method [IEEE Transactions on Biomedical Engineering, 49(12), 2002] to a scheme in which the estimation of the corneal limbus from a placido-disk videokeratoscopic images is performed in a more accurate manner. This has been achieved with the help of a bootstrap technique that calculates the bias of the corneal limbus estimator. Results Corneal limbus estimation was performed on 16 subjects (32 eyes) from a variety of ethnic backgrounds. The refined technique for estimating corneal limbus that includes the bootstrap estimation step showed better accuracy (<50 microns) when compared to the previous method. The accuracy is relative to a manual computer aided technique in which an experienced operator selects points in an image that lie on the limbus. Conclusions The proposed algorithm for estimating corneal limbus can add important clinical value to the output of the placido disk videokeratoscopes. The technique can also be used for robust estimation of the pupil diameter.