Rotating Scheimpflug Imaging Indices in Different Grades of Keratoconus

Purpose. To evaluate accuracy of various Keratoconus (KC) screening indices, in relation to Topographic Keratoconus (TKC) grading. Setting. Al Watany Eye Hospital, Cairo, Egypt. Methods. Data of 103 normal (group 1) and 73 KC eyes (group 2), imaged by Pentacam (branded as Allegro Oculyzer), were analysed. Group 2 was divided into 2a: 14 eyes (TKC = 1, early KC), 2b: 25 eyes (TKC = 1 to 2 or 2, moderate KC), and 2c: 34 eyes (TKC = 2 to 3 up to 4, severe KC). Participants were followed up for six years to confirm diagnosis. Area under the receiver operating characteristic curve (AUROC) was calculated for evaluated curvature, elevation, and pachymetry indices with various reference shapes at different diameters. Results. When comparing normal to KC eyes, ten indices had significantly higher AUROC. Only five of them had significantly higher AUROC in early KC compared to normal corneas: Pachymetry Progression Index- (PPI-) Maximum (Max), Ambrósio's Relational Thickness- (ART-) Max, PPI-Max minus PPI-Minimum (Min), central corneal thickness (CCT), and diagonal decentration of thinnest point from the apex (AUROC = 0.690, 0.690, 0.687, 0.683, and 0.674, resp.). Conclusion. Generally, ten pachymetry and elevation-based indices had significantly higher AUROC. Five indices had statistically significant high AUROC when comparing early KC to normal corneas.


Introduction
The hallmark characteristic of Keratoconus (KC) is the development of localized, cone-shaped ectasia accompanied by corneal thinning in the area of the cone. This leads to irregular astigmatism and steeper corneal curvature over the area of the cone [1]. The early detection of these changes is usually based on Placido-based corneal topography and rotating Scheimpflug camera corneal tomography. However, there are confounding conditions that can imply topographic and tomographic corneal changes resembling KC, such as corneal scarring, hyperopic refractive surgery, contact lens-induced warpage [2], and even normal variants of topographic curvature patterns.
There are several tomographic criteria for KC diagnosis. They can be divided into three main subgroups: curvaturebased, elevation-based, and pachymetry-based. Unfortunately, none of them is 100% sensitive and specific. Some surface is the most sensitive reference body to compare KC to normal corneas [8][9][10].
More recently, there have been other proposed indices; one of the most valuable KC indices is the Ambrósio's Relational Thickness (ART) [11].
This study aims at comparing the accuracy of each of the tested indices at various best-fit reference surfaces, in cohorts of KC and normal corneal cases who were followed up for six years, and to evaluate the accuracy of various indices in relation to Topographic KC (TKC) grading.
To the best of our knowledge, our study is innovative in evaluating the accuracy of KC indices using different reference surfaces of various diameters when applicable. Moreover, novelty lies in comparing the tested indices against time evidence, through our six-year follow-up interval (stability even after LASIK surgeries of normal corneas and spontaneous progression over years in KC cases). Time evidence is a definite proof of diagnosis not relying on subjective assessment.

Patients and Methods
A retrospective study was conducted using the data from consecutive patients' files that were imaged in the time interval between June 2008 and December 2009, using the Pentacam branded as Allegro Oculyzer (WaveLight, GmbH, Erlangen, Germany) [12] with software version 1.16r12 at Al Watany (National) Eye Hospital, Cairo, Egypt. Patients with any detected corneal pathology other than KC and patients with history of ocular surgery, contact lens wear during the previous two weeks, or narrow palpebral fissure impeding reliable imaging were excluded from the study. Any patient with questionable diagnosis at the time of initial Pentacam imaging was also excluded. Moreover, the patients were followed up until December 2015 to confirm the diagnosis and hence to confirm and evaluate the results.
Every eye was scanned at least thrice by the Allegro Oculyzer, in a dark room and according to the recommendations stated in the device manual. Each scan included 25 Scheimpflug images. Despite good repeatability, data were collected from the most reliable scan as stated by the "QS" pop-up box (i.e., largest analysed area, valid data percent, and good alignment). The data were collected from the automatically calculated indices or generated by manually changing the reference surface shape (BFS or BFTE) and calculation area (7, 8, or 9 mm) and getting the elevation values on mouse click. The investigated indices were as follows.
Curvature indices are as follows: (iv) Maximum elevation, minimum elevation, and the difference between them (in m) from the 8 and 9 mm anterior BFS.
(v) Maximum elevation, minimum elevation, and the difference between them (in m) from the 7, 8 and 9 mm posterior BFS.
(vi) Maximum elevation, minimum elevation, and the difference between them (in m) from the 7, 8, and 9 mm posterior BFTE.
Pachymetry indices are as follows: (i) Pachymetry at corneal apex, corneal center, and thinnest point (in m).
(iv) Horizontal and vertical decentration of the thinnest point from the apex and the resultant diagonal decentration [=√(squared horizontal decentration + squared vertical decentration) (in mm)].
The study adhered to the Tenets of the Declaration of Helsinki and to the local ethics standards.   The following tests were performed: calculation of the mean, standard deviation (SD), unpaired -test, sensitivity, specificity at different cut-off values, likelihood ratios, and area under the receiver operating characteristic curve (AUROC). AUROCs are compared using the DeLong method [13]. Values were considered statistically significant if value was less than 0.05. The mean, SD, and 95% confidence interval (CI) of each parameter are listed in Table 1. Table 2 represents the AUROC of all tested indices when comparing normal corneas to all other grades of KC collectively, and Table 3 represents the ten indices with the highest AUROC. The AUROCs of these ten indices were statistically noninferior to each other ( > 0.05). Meanwhile, all other indices had statistically inferior AUROC to that of at least one of these ten indices. It is worth mentioning that none of the curvature indices was among these ten indices.

Cases
Journal of Ophthalmology 5        Table 4.

Discussion
Placido-based systems rely on the analysis of a reflected image. This cannot provide data from the posterior corneal surface [6]. Additionally, without information about the posterior corneal surface, complete pachymetric evaluation is not possible [14]. Ultrasonic central pachymetry measures one point, which is not necessarily the thinnest point, and does not reflect the overall thickness profile [15]. Moreover, the posterior corneal surface is appreciated as a sensitive indicator of corneal ectasia and can often be abnormal in spite of a still normal anterior corneal surface. While the corneal refractive power is largely determined by the anterior surface, the evaluation of anatomical properties of the cornea is at least equally dependent upon both surfaces [14]. Therefore, corneal tomography parameters assessment appeared to be essential in detecting KC.
With the appearance of many KC indices and suggestion of various best-fit reference surfaces [6], the goal of our study was to evaluate the accuracy (including both sensitivity and specificity) of such indices with the use of different reference surfaces, at different diameters, and correlate these parameters to TKC grading. Some previous studies used cyclic evidence that stated that a certain index was accurate based on differentiating KC from normal cornea using certain criteria closely related to this same index. Our study adds the long follow-up (6 years) as a more robust evidence. If a cornea was falsely diagnosed as normal and underwent LASIK, it would develop ectasia in such a long follow-up period. On the other side, a KC diagnosis can be confirmed if it spontaneously progressed over time.
The inclusion of every single possible index is out of the scope of this study. The included indices are those mentioned individually in other studies or those we thought they might be significant. On evaluating BFTE and smaller best-fit surface (7 mm), we focused on the posterior rather than the anterior corneal surface, being the site for primary subclinical tomographic changes, preceding the anterior surface [16][17][18][19].
Curvature-based indices derived from both anterior and posterior surfaces were evaluated and analysed. According to the ROC curves, all evaluated curvature indices had statistically significant less AUROC for diagnosing KC than most of elevation and pachymetry indices. Previous reports demonstrated similar results [20][21][22]. Despite being the least significant in diagnosing KC, they have been considered as important criteria in KC and after LASIK ectasia follow-up [23,24].
In this study, where we only analysed data from good quality scans with corneal surface area more than 9 mm, analysis of elevation indices showed that posterior surface elevation (PE) from 8 mm BFS and anterior surface elevation (AE) from 9 mm BFS had the highest accuracy, with AUROC (0.979) for both, followed by PE from 7 and 9 mm BFS (AUROC = 0.977 and 0.971, resp.). AE and PE from reference best-fit surfaces did not differ in their accuracy with various diameters (7, 8, and 9 mm). Correia et al. reported that PE from both 8 mm BFS and BFTE had highest AUROC (0.983 and 0.986, resp.) [20]. de Sanctis et al. [25] evaluated the sensitivity and specificity of PE from 9 mm BFS in discriminating normal corneas from KC. The AUROC analyses showed high overall predictive accuracy of PE for KC (AUROC 0.99). They concluded that PE very effectively discriminates KC from normal corneas, but its efficacy is lower for subclinical KC, and thus data concerning PE should not be used alone to stratify patients with this condition.
We enrolled in our work the elevation indices of the thinnest point from both BFS and BFTE. The latter was evaluated in literature and showed higher predictive accuracy in diagnosing KC and forme fruste KC compared to BFS, especially with a fixed eccentricity. A possible explanation for its diagnostic superiority in ectatic corneas is that its central vaulting can determine whether the corneal pattern is associated with an atypical condition, a true corneal disease, or an artifact of alignment or processing [16,20].
According to our study results, parameters that had the highest AUROC for diagnosing KC were the PPI-Max at cutoff value of >1.4 and the ART-Max at cut-off value of ≤412 (0.987 for both). These results were comparable to those of Ambrósio Jr. et al. [1] who reported that the most sensitive 8 Journal of Ophthalmology parameters were ART-Avg and ART-Max (AUROC = 0.987 and 0.983, resp.).
Our study investigated the accuracy of various tomographic indices in relation to TKC grading [26]. We found that most of the tested indices had significantly different values between each of grades 2 to 4 (moderate and/or severe) KC and normal corneas. In 2012, Ishii and his coworkers [27] investigated the severity of KC in terms of corneal elevation differences, and they correlated the data with Amsler-Krumeich classification of KC. They observed that cases of higher elevation differences in the anterior and posterior corneal surfaces were staged higher in Amsler-Krumeich classification ( < 0.001) and concluded that anterior and posterior corneal surface height data obtained by elevationbased tomography provide useful information in improving KC diagnostic accuracy and in grading the severity of KC. Their results are comparable to ours in the accuracy of elevation indices; however, they did not furtherly investigate other indices.
Furthermore, Kamiya and his colleagues [28] evaluated corneal elevation, pachymetry, and keratometry in KC eyes according to the clinical stage of the disease, with respect to each stage of Amsler-Krumeich classifications. They detected that PE (0.980) and AE (0.977) showed the highest AUROCs, followed by max value (0.941), percentage thickness increase (PTI) 2 mm (0.931), PTI 4 mm (0.927), and PPI-Max (0.927). Their results are highly comparable to ours in AUROC of indices. They observed that AE and PE measurements tended to have a higher accuracy at the earlier stages of KC, when compared to other studied parameters; hence, they concluded that elevation difference measurements might provide useful information for improving the diagnostic accuracy of KC, especially in the early stage. These results contradict ours, where we found that the indices with statistically significant high AUROC when comparing grade 1 KC to normal corneas were PPI-Max, ART-Max, PPI-Max minus PPI-Min, central corneal thickness, and diagonal decentration of the thinnest point from the apex.
Our findings justify the modifications in the newer software versions that incorporated "ART-Max" and "diagonal decentration of the thinnest point from the apex," being sensitive indices for early KC detection. As regards the hardware changes, we also recommend similar studies on Pentacam HR, because the current one was on Allegro Oculyzer I Pentacam. Randleman et al. [29] reported that, for refractive surgical screening, regular and high-resolution Scheimpflug imaging devices generated different objective values and the two devices are not interchangeable.

Conclusion
The accuracy of different curvature, elevation, and pachymetry-based indices using various reference shapes of different diameters was evaluated, in relation to TKC grading. To the best of our knowledge, this is the first study to confirm clinical and topographical observations by a long follow-up period of 6 years to ascertain the diagnosis by time evidence. Generally, to discriminate all KC grades from normal, ten pachymetry and elevation-based indices were significantly more accurate than other indices, having higher AUROC. Five of them had statistically significant high AUROC when comparing early KC to normal corneas.

Competing Interests
There are no competing interests of any of the authors with any establishment having a relation to this present work.