Journal of Refractive Surgery

The articles prior to January 2013 are part of the back file collection and are not available with a current paid subscription. To access the article, you may purchase it or purchase the complete back file collection here

DIAGNOSTIC TECHNIQUES 

Screening Patients With the Corneal Navigator

Stephen D Klyce, PhD; Michael D Karon, PhD; Michael K Smolek, PhD

Abstract

ABSTRACT

PURPOSE: To present a corneal topography screening device for the detection of corneal ectasias and various refractive procedures based on corneal topography patterns.

METHODS: A database of corneal topography patterns were analyzed and used to "train" a neural network on nine different corneal topography patterns using nineteen corneal topography indices of corneal shape and power.

RESULTS: Sample normal and corneal topographies were recognized correctly.

CONCLUSIONS: The use of the corneal navigator to screen various corneal topographies aids clinical diagnosis. [J Refract Surg. 2005;21(Suppl):S617-S622.]

Abstract

ABSTRACT

PURPOSE: To present a corneal topography screening device for the detection of corneal ectasias and various refractive procedures based on corneal topography patterns.

METHODS: A database of corneal topography patterns were analyzed and used to "train" a neural network on nine different corneal topography patterns using nineteen corneal topography indices of corneal shape and power.

RESULTS: Sample normal and corneal topographies were recognized correctly.

CONCLUSIONS: The use of the corneal navigator to screen various corneal topographies aids clinical diagnosis. [J Refract Surg. 2005;21(Suppl):S617-S622.]

Corneal topographic analysis is commonly used as a screening device for determining progressive ectatic corneal dystrophies such as keratoconus and pellucid marginal degeneration.115 As a screening device to determine the suitability of candidates for refractive surgery or as a diagnostic tool for routine clinical use, corneal topographic analysis is the standard of care. Corneal topographic analysis can prevent pathologic corneas from undergoing refractive surgery and the ensuing complications. The clinical decision for differentiating between normal corneal asymmetry and asymmetry that is suspicious for corneal disease can be challenging. For example, many corneas may have inferior steepening patterns preoperatively that marginally impinge the modified Rabinowitz inferior-superior test3 (having a vertical power gradient on an axial map >1.4 diopters [D] over a 6-mm distance), yet after refractive surgery go on to develop iatrogenic keratectasia (Fig 1). Additionally, forme fruste keratoconus often is only detectable by corneal topography.

Determining normal and abnormal corneal topography patterns based on fine subtleties of steepening changes can be challenging. This necessitates a more conservative approach to screening than in the past. The use of artificial intelligence schemes to automatically detect abnormal corneal topography as an aid in screening may be valuable.

MATERIALS AND METHODS

The NIDEK OPD-Scan corneal topography (NIDEK, Gamagori, Japan) and the Magellan Mapper (NIDEK) were used to measure all corneal topography data. Corneal topographic examination and diagnoses were provided by five expert clinicians. Corneal topographies were characterized as normal, astigmats, keratoconus suspect, keratoconus, pellucid marginal degeneration, post-penetrating keratoplasty, myopic refractive surgery, hyperopic refractive surgery, and other. To overcome the differences in mire patterns among the topographers, after converting the corneal power maps to height maps, a Fourier analysis filter was applied to transform the data to the frequency domain. This allowed the different Placido devices to yield the same information. From this information indexing schemes were developed to differentiate among the different classifications of corneal topography. One neural network was developed for each category of corneal topography. Nineteen indices of corneal shape and power were analyzed to determine normal, suspect, and abnormal values, which were color-coded green, yellow, and red, respectively. Abnormal indices are yellow in color and represent values that are 2 to 3 standard deviations from the norm. Abnormal indices are red in color and represent values that are >3 standard deviations from the norm.

Figure 1. A) Pre- and B) postoperative axial corneal topography maps of a patient who developed iatrogenic keratectasia.Figure 2. Representation of a normal corneal topography with the NIDEK Corneal Navigator, an examination made with the wide mire NIDEK OPD-Scan.

Figure 1. A) Pre- and B) postoperative axial corneal topography maps of a patient who developed iatrogenic keratectasia.

Figure 2. Representation of a normal corneal topography with the NIDEK Corneal Navigator, an examination made with the wide mire NIDEK OPD-Scan.

RESULTS

Normal corneal topography appears on this automated classification system as a percent similarity to the normal corneal examinations that the neural network was trained on (Fig 2). The various indices of corneal shape and power are color-coded to alert the clinician to normal (green), suspect (yellow), or abnormal (red) values. Figure 2, for example, shows an average central corneal power of 43.29 D and a corneal eccentricity index of 0.41; both of these indices are within normal limits in the examination presented.

Figure 3. Sample corneal topography classification of cornea that had undergone myopic refractive surgery (obtained with the Nidek OPD-Scan).

Figure 3. Sample corneal topography classification of cornea that had undergone myopic refractive surgery (obtained with the Nidek OPD-Scan).

Figure 3 represents a pattern of an eye that has undergone myopic refractive surgery. The axial topography maps show central flattening surrounded by a steeper periphery. The classifier denotes that a 99% similarity to myopic refractive surgery is present. Abnormal index values are highlighted in yellow and red such as the average central corneal power of 39.94 D and a corneal eccentricity index of ?0.69, denoting an oblate cornea. An example of keratoconus is classified in Figure 4, showing inferior steepening on axial topography map and a similarity index of 86% and a keratoconus severity index of 35.1%. Although both average central corneal power and corneal eccentricity index are within normal limits in this mild keratoconus, a number of the other variables are elevated, signaling to the keratoconus network a positive response.

Pellucid marginal degeneration, a corneal ectasia not as common as keratoconus, is shown in Figure 5. An inferior steepening pattern is seen on the axial map and only 5 of the 19 indices of the classifier are within normal limits. The hallmarks of pellucid marginal degeneration are the central "negative" low power bow tie and the inferior arcuate elevated power in the form of a claw.

Figure 6 represents an ambiguous case with 0.48 D of cylinder along with a mild inferior steepening that has a pattern with 23.8% similarity to keratoconus suspect topographies used to "train" the neural network.

Figure 7 represents the corneal topography of a patient who underwent a treatment for myopia using a prolate ablation algorithm with the NIDEK EC-CX excimer laser. The postoperative cornea is prolate as seen on the axial map and is classified as normal with only 1 index (slightly lower than normal central power) of 19 considered suspect (Fig 7A). The potential visual acuity of this cornea was excellent (20/13) and visual inspection of the topography using the universal standard fixed scale is unable to detect the surgery. It was necessary to use a topographic difference map (Fig 7B) that subtracts the postoperative topography from the preoperative topography to reveal the tissue subtraction that has occurred.

DISCUSSION

Developing an automated classification system to detect corneal pathologies and corneas that have undergone various surgical procedures can be a useful adjunct to clinical diagnosis. Refractive surgery on eyes with clinically undetected ectatic degenerative conditions can lead to iatrogenic ectasia over time. A retrospective study by Randelman et al16 found that every case that developed keratectasia after LASIK was due either to an ablation that violated the "250 micron" rule or exhibited topographic abnormalities preoperatively that were similar to keratoconus or keratoconus suspect patterns. A retrospective analysis on eyes after photorefractive keratectomy (PRK) that went on to develop keratectasia all exhibited topographies that were similar to forme fruste keratoconus (Roberto Zaldivar, MD, personal communication, 2003). Currently, LASIK is contraindicated in patients who are suspects for corneal ectatic degenerative conditions; however, PRK is still considered a viable option by some surgeons. Preliminary data analysis by Zaldivar (personal communication, 2003) indicates that opting to correct refractive errors with PRK rather than LASIK may only delay the onset of postoperative keratectasia from 9 to 18 months with LASIK to 4 or 5 years with PRK in those corneas with suspect ectatic pathology.

Figure 4. Sample corneal topography classification of keratoconus using the fine mire NIDEK Magellan topographer.Figure 5. Sample corneal topography classification of pellucid marginal degeneration using the NIDEK Magellan topographer.

Figure 4. Sample corneal topography classification of keratoconus using the fine mire NIDEK Magellan topographer.

Figure 5. Sample corneal topography classification of pellucid marginal degeneration using the NIDEK Magellan topographer.

Figure 6. Simultaneous classification of corneal astigmatism and keratoconus suspect.Figure 7. A) Corneal topography classification of a postoperative prolate ablation using the NIDEK EC-CX excimer laser. Although the cornea underwent a myopic ablation, a prolate shape was maintained and the Corneal Navigator classified this as normal. B) A difference map (center right) reveals the area of tissue subtraction that occurred between the preoperative shape (lower left) and the postoperative shape (upper left) of a cornea that was treated with the NIDEK EC-CX excimer laser.

Figure 6. Simultaneous classification of corneal astigmatism and keratoconus suspect.

Figure 7. A) Corneal topography classification of a postoperative prolate ablation using the NIDEK EC-CX excimer laser. Although the cornea underwent a myopic ablation, a prolate shape was maintained and the Corneal Navigator classified this as normal. B) A difference map (center right) reveals the area of tissue subtraction that occurred between the preoperative shape (lower left) and the postoperative shape (upper left) of a cornea that was treated with the NIDEK EC-CX excimer laser.

Mild to advanced keratoconus and pellucid marginal degeneration are easily detectable with corneal topography. These stages of clinical keratoconus exhibit an inferior steepening, producing marked topographic asymmetry. Mild to advanced pellucid marginal degeneration is characterized by against-the-rule astigmatism producing a central "negative" bow tie and a lobster claw or "C-shaped" arcuate band of raised corneal power. These stages are generally accompanied by regional corneal thinning as well. However, more subtle variations in power (see Fig 6) become more of a clinical challenge to diagnose. Forme fruste keratoconus often is only detectable using corneal topography, as topographic changes appear prior to any presenting signs such as Vogt's striae or a Fleischer ring on slitlamp examination. Hence, the development of a screening method such as the Corneal Navigator aids in the objective evaluation of the topographic patterns. By characterizing and classifying a variety of topography patterns, the clinician will be able to determine whether the cornea is nascent or had undergone surgery. As the popularity of intraocular lenses increases, the surgeon will need to determine whether a patient underwent myopic or hyperopic surgery to compensate for the correct implant power. Whether a patient underwent conventional myopic or hyperopic ablation is easily determined by looking at the corneal topography patterns. Current laser algorithms with certain lasers create a prolate rather than oblate corneal shape; these topographies are currently classified as normal by this system primarily due to the lack of available data sets for such treatments. Although this is a benefit to the patient that will reduce night vision complaints in particular, we await a larger number of examples of such treatments to retrain the Corneal Navigator to recognize this refined surgical alteration.

The current screening system is successful in recognizing corneal topography patterns for specific corneal ectasias and refractive surgeries. Implementation and refinement of such systems will aid the clinical diagnosis and management of anterior segment patients.

REFERENCES

1. Rowsey JJ, Reynolds AE, Brown R. Corneal topography. Corneascope. Arch Ophthalmol. 1981;99:1093-1100.

2. Dingeldein SA, Klyce SD. Imaging of the cornea. Cornea. 1988;7:170-182.

3. Rabinowitz YS, McDonnell PJ. Computer -assisted corneal topography in keratoconus. Refract Corneal Surg. 1989;5:400-408.

4. McDonnell PJ. Current applications of the Corneal Modeling System. Refract Corneal Surg. 1991;7:87-91.

5. Wilson SE, Klyce SD. Screening for corneal topographic abnormalities before refractive surgery. Ophthalmology. 1994; 101: 14 7-152.

6. Maeda N, Klyce SD, Smolek MK, Thompson HW. Automated keratoconus screening with corneal topography analysis. Invest Ophthalmol Vis Sci. 1994;35:2749-2757.

7. Hjortdal JO, Erdmann L, Bek T. Fourier analysis of video-keratographic data. A tool for separation of spherical, regular astigmatic and irregular astigmatic corneal power components. Ophthalmic Physiol Opt. 1995;15:171-185.

8. Maeda N, Klyce SD, Smolek MK. Neural network classification of corneal topography. Preliminary demonstration. Invest Ophthalmol Vis Sci. 1995;36:1327-1335.

9. Maeda N, Klyce SD, Smolek MK. Comparison of methods for detecting keratoconus using videokeratography. Arch Ophthalmol. 1995;113:870-874.

10. Kalin NS, Maeda N, Klyce SD, Hargrave S, Wilson SE. Automated topographic screening for keratoconus in refractive surgery candidates. CLAOJ. 1996;22:164-167.

11. Smolek MK, Klyce SD. Current keratoconus detection methods compared with a neural network approach. Invest Ophthalmol Vis Sci. 1997;38:2290-2299.

12. Rabinowitz YS, Rasheed K. KIS A% index: a quantitative videokeratography algorithm embodying minimal topographic criteria for diagnosing keratoconus. J Cataract Refract Surg. 1999;25:1327-1335.

13. Chastang PJ, Borderie VM, Carvajal-Gonz?lez S, Rostene W, Laroche L. Automated keratoconus detection using the Eye Sy s videokeratoscope. J Cataract Refract Surg. 2000;26:675-683.

14. Ambrosio R Jr, Klyce SD, Wilson SE. Corneal topographic and pachymetric screening of keratorefractive patients. J Refract Surg. 2003;19:24-29.

15. Tomidokoro A, OshikaT. Quantitative assessment of videokeratography data using fourier series harmonic analysis. Cornea. 2004;23:S71-S77.

16. Randleman JB, Russell B, Ward MA, Thompson KP, Stulting RD. Risk factors and prognosis for corneal ectasia after LASIK. Ophthalmology. 2003;110:267-275.

10.3928/1081-597X-20050902-12

Sign up to receive

Journal E-contents