Use of machine learning to achieve keratoconus detection skills of a corneal expert

PURPOSE

To construct an automatic machine-learning derived algorithm discriminating between normal corneas and suspect irregular or keratoconic corneas.

METHODS

A total of 8526 corneal tomography images of 4904 eyes obtained between November 2010 and July 2017 using a combined Scheimpflug/Placido tomographer were retrospectively evaluated. Each image was evaluated for acquisition quality and was labeled as normal, suspect irregular or keratoconic by a cornea specialist. Two algorithms were built. The first was based on 94 instrument-derived output parameters, and the second integrated keratoconus prediction indices of the device with the 94 instrument-derived output parameters. Both models were compared with the tomographer's keratoconus detection algorithms. Out of the 8526 images evaluated, 7104 images of 3787 eyes had sufficient acquisition quality. Of those, 5904 examinations were randomly chosen for construction of the models using the random forest algorithm. The models were then validated using the remaining 1200 examinations.

RESULTS

Both RF algorithms had a larger AUC compared with any of the tomographer's KC detection algorithms (p < 10^-9). The first constructed model had 90.2% accuracy, sensitivity of 94.2%, and specificity of 89.6% (Youden 0.838). Calculated AUC was 0.964. The second model had 91.5% accuracy, sensitivity of 94.7%, and specificity of 89.8% (Youden 0.846). Calculated AUC was 0.969.

CONCLUSION

Using the RF machine-learning algorithm, accuracy of discrimination between normal, suspect irregular and keratoconic corneas approximates that of an experienced corneal expert. Applying machine learning to corneal tomography can facilitate keratoconus screening in large populations as well as off-site screening of refractive surgery candidates.

Dual-Scheimpflug-Placido-disc corneal analysis in cats

OBJECTIVE

To evaluate the corneal topography and tomography of cats using a dual Scheimpflug and Placido disc device.

ANIMAL STUDIED

Sixteen mixed-breed cats (32 eyes).

PROCEDURE(S)

Cats were sedated using meperidine (4.3 mg/kg/IM) and xylazine (0.5 mg/kg/IM), and evaluated using the software Galilei™. Values assessed included anterior and posterior corneal curvature, total corneal power, and corneal thickness.

RESULTS

Ten males and six females aged between 18 and 48 months were evaluated. The simulated keratometry (SimK) was 38.37 ± 0.83D, mean posterior keratometry -4.89 ± 0.34D, and mean total corneal power 38.04 ± 0.94D. The mean central corneal thickness (CCT) was 617.34 ± 53.38 μm with a mean thinnest point of 583.74 ± 60.60 μm.

CONCLUSION

The Galilei™ examination was feasible in cats, enabling a detailed study of the cornea. To the best of our knowledge, this is the first study evaluating the corneal topography and tomography of cats using a device that integrates data from a dual Scheimpflug and Placido disc system.

Simplifying and understanding various topographic indices for keratoconus using Scheimpflug based topographers

Keratoconus (KC) is a progressive ectatic corneal disorder. There are multiple topographic devices and their varied indices used for diagnosis, detecting progression, and deciding management. It is important to understand the repeatablility, intra- test variabililty, and comparability amongst various topographic devices. The Scheimpflug camera-based devices, such as the Pentacam (Oculus, Wetzlar, Germany), Galilei (Ziemer, Biel, Switzerland), and Sirius (Costruzione Strumenti Oftalmici, Florence, Italy) are known to assist in the detection of early keratoconus and subclinical keratoconus. This article reviews the various Scheimpflug camera-based devices in depth, addressing their different indices, diagnostic accuracy, repeatability, and agreement and identifying the strongest parameter of each device. It will guide the practicing clinician by giving practical tips for decision making in the diagnosis and management of keratoconus.

Comparison of anterior segment changes after femtosecond laser LASIK and SMILE using a dual rotating Scheimpflug analyzer

Background

To compare the changes in the anterior segment after femtosecond laser in situ keratomileusis (FS-LASIK) and small incision lenticule extraction (SMILE) using a dual rotating Scheimpflug (DRS) analyzer (Galilei®; Ziemer Ophthalmology, Port, Switzerland).

Methods

A total of 218 eyes of 109 patients who underwent FS-LASIK or SMILE for myopic correction were retrospectively studied. Ninety-eight eyes of 49 patients who underwent FS-LASIK were compared to 120 eyes of 60 patients treated with SMILE. A DRS analyzer was used for preoperative and 6-month postoperative anterior segment analyses. Measured variables included the central corneal thickness (CCT), anterior chamber depth (ACD), anterior and posterior keratometry (K), anterior and posterior best-fit sphere radius, and maximum posterior elevation (MPE).

Results

After the procedure, the amount of CCT decrease was higher in the SMILE group than in the FS-LASIK group, but it was not statistically significant. The MPE was significantly increased after both procedures (p < 0.001 and p = 0.001 in the FS-LASIK and SMILE groups, respectively), with the amount of elevation being higher after FS-LASIK than after SMILE even though it was not statistically significant. And there was a significant change in the steep and average posterior K in the FS-LASIK group (p = 0.006 and 0.001, respectively), but not in the SMILE group.

Conclusions

Regarding changes in the MPE and posterior K, changes in the posterior corneal surface were greater after FS-LASIK than after SMILE.

Trial registration

The trial registration number: KCT0003628. Date of registration: 15 March 2019.

Comparison of using Galilei Dual Scheimpflug Analyzer G4 and Barrett formula in predicting low cylinder preoperatively for cataract surgeries

PURPOSE

To compare corneal astigmatism estimation from Barrett toric calculator, with measurement from Galilei Dual Scheimpflug Analyzer G4 in low corneal cylinder patients.

METHODS

Preoperative corneal astigmatism was measured using Auto Kerato-Refractometer (AutoKM), IOL Master, and Galilei G4 (combined Placido-dual Scheimpflug analyzer) and was processed by Barrett toric calculator with measurements obtained from Auto Kerato-Refractometer and from IOL Master. A total of 42 eyes undergoing cataract surgery with nontoric intraocular lens implantation were included. Corneal astigmatism was calculated based on manifest refractive astigmatism with implications of surgically induced astigmatism. Errors in predicted residual astigmatism were calculated by the difference between postoperative manifest cylindrical refractive error and preoperative corneal cylinder using vector analysis.

RESULTS

Centroid error in predicted residual astigmatism was with-the-rule 0.36 D for AutoKM and 0.48 D for IOL Master, was lower at 0.24 D for the Barrett-IOL Master, and was lowest at 0.21 D for the Barrett-AutoKM (p < .001). The Galilei G4 demonstrated the highest centroid error for SimK (0.53 D) and lower for total corneal power (0.49 D). The Barrett toric calculator obtained the lowest median absolute error in predicted residual astigmatism for AutoKM (0.43 D) and IOL Master (0.54 D). The Barrett-IOL Master demonstrated that 61% and 76% of eyes were within 0.50 and 0.75 D of the predicted residual astigmatism, respectively.

CONCLUSION

The Barrett-IOL Master had more accurate prediction of residual astigmatism for low astigmatism eyes before cataract surgery compared to Galilei Dual Scheimpflug Analyzer G4 in this study.

Advances in Biomechanical Parameters for Screening of Refractive Surgery Candidates: A Review of the Literature, Part III

Corneal biomechanical properties have garnered significant interest in their relation to the development of ectatic corneal disease. Alongside the advent of corneal tomography and Scheimpflug imaging such as Pentacam and Galilei, there have been advances in assessing the cornea based on its biomechanical characteristics. Though the aforementioned imaging systems are highly capable of identifying morphologic abnormalities, they cannot assess mechanical stability of the cornea. This article, in contrast to Parts I and II of this article series, will focus on in vivo corneal biomechanical imaging systems. The two most readily available commercial systems include the Corvis ST and the Ocular Response Analyzer. Both of these systems aimed to characterize corneal biomechanics via distinct measurements. While in Parts I and II of this article series the authors focused on elevation, pachymetric, and keratometric data, the purpose of this article was to summarize biomechanical parameters and their clinical use in screening refractive surgery candidates. Moreover, this article explores biomechanical decompensation and its role in the development of corneal ectasia and keratoconus. There is a focus on the diagnostic accuracy of biomechanical indices in the identification of diseases such as keratoconus that may preclude a patient from undergoing refractive surgery.

Agreement between Keratometric readings by VERION image guided System, Galilei G4 and Pentacam

Objective:

To study Agreement between Keratometric readings by VERION image guided System, Galilei G4 and Pentacam.

Methods:

The quasi experimental study was conducted at Armed Forces Institute of Ophthalmology, Rawalpindi, Pakistan from August 2016 to December 2016. Twenty five patients fulfilling the inclusion criteria participated in the study. All Patients were subjected to Keratometric assessment using Galilei G4 Dual Scheimpflug analyzer (Ziemer, Switzerland), Wavelight Oculyzer II (Pentacam, Germany) and Verion image guided system (Alcon). Steep and flat meridian and diopter of astigmatism by three systems were recorded and endorsed. All readings were taken by the same observer. Statistical Program for Social Sciences (SPSS) version 22.0 was used for statistical analysis. Results analyzed for significance by t-tests and Interclass correlation analysis. In t tests, P values of <0.05 was considered statistically significant while interclass coefficient of >0.7 was considered acceptable.

Results:

Fifty eyes of twenty-five patients (22 male, 28 female) with mean age of 29.50 ± 3.46 years were studied. Flat K, steep K and dioptric power of astigmatism were measured with verion, pentacam and Galilei G4. Interclass correlation analysis showed agreement between individual variables measured by the three devices, while one sample t test showed no significant difference between dioptric power of astigmatism between Verion-Pentacam and Verion- Galilei group. (p 0.178 for former and 0.622 for later group).

Conclusion:

Verion image guided system is comparable to other instruments used currently for keratometry. Verion can be interchangeably used with Pentacam and Galilee G4 in assessing corneal astigmatism.

Comparison between Two Scheimpflug Anterior Segment Analyzers

Purpose:

To compare the anterior segment indices measured by two Scheimpflug camera machines; Galilei and Pentacam.

Methods:

In this observational case series, the anterior segment indices of myopic healthy subjects seeking for refractive surgery were measured by Pentacam and Galilei on the same day. Analyzed parameters were anterior and posterior best fit spheres (BFS), axial curvature, true corneal power, central corneal thickness (CCT), anterior chamber (AC) depth, AC volume, AC angle, and pupil diameter.

Results:

This study included 176 eyes of 88 participants. Mean radius of the anterior BFS was 7.79 ± 0.34 mm versus 7.75 ± 0.39 mm measured by Pentacam and Galilei, respectively (r = 0.877, P < 0.001). Corresponding values for the mean radius of posterior BFS were 6.42 ± 0.32 and 6.47 ± 0.38 mm, respectively (r = 0.879, P < 0.001). Anterior corneal mean power was 43.8 ± 1.9 diopters (D) with Pentacam and 43.8 ± 2.4 D with Galilei (r = 0.905,P < 0.001). Posterior corneal mean power was measured − 6.3 ± 0.3 and − 6.3 ± 0.4 D using Pentacam and Galilei, respectively (r = 0.873, P < 0.001). True corneal power was 43.9 ± 1.9 D with Pentacam and 43.5 ± 2.3 D with Galilei (r = 0.909, P < 0.001). CCT was 537 ± 44 and 553 ± 51 μm measured by Pentacam and Galilei, respectively (r = 0.796, P < 0.001). AC depth measurements using Pentacam and Galilei were 3.29 ± 0.4 and 3.3 ± 0.38 mm (P < 0.001), respectively; AC volume was 207 ± 50 and 129 ± 39 mm³≥ (P = 0.004), and AC angle was 39.7 ± 9.2 and 54.2 ± 5.2 degrees (P = 0.051), respectively. Average pupil diameter was measured 3.91 ± 1.77 mm by Pentacam and 3.34 ± 0.89 mm by Galilei (P = 0.018).

Conclusions:

There was a significant correlation between the Pentacam and Galilei in all measured parameters except AC angle, AC volume, and average pupil diameter.

Repeatability and Reproducibility of Anterior Segment Measurements in Normal Eyes Using Dual Scheimpflug Analyzer

OBJECTIVES

To assess the repeatability and reproducibility of anterior segment measurements including aberrometric measurements provided by a dual Scheimpflug analyzer (Galilei) system in normal eyes.

MATERIALS AND METHODS

Three repeated consecutive measurements were taken by two independent examiners. The following were evaluated: total corneal power and posterior corneal power, corneal higher-order wavefront aberrations (6.0 mm pupil), pachymetry at the central, paracentral, and peripheral zones, and anterior chamber depth (ACD). Repeatability was assessed by calculating the within-subject standard deviation, precision, repeatability, and intraclass correlation coefficient (ICC). Bland-Altman analysis was used for assessing reproducibility.

RESULTS

Thirty eyes of 30 patients were included. The best ICC values were for corneal pachymetry and ACD. For both observers, acceptable ICC was also achieved for the other parameters, the only exceptions being posterior corneal astigmatism and total high order aberration. The 95% LoA (Limits of Agreement) values for all measurements showed small variability between the two examiners.

CONCLUSION

The Galilei system provided reliable measurements of anterior segment parameters. Therefore, the instrument can be confidently used for routine clinical use and research purposes.

Central and peripheral corneal thickness measurement in normal and keratoconic eyes using three corneal pachymeters

Purpose:

To assess the agreement among ultrasonic pachymetry, the Galilei dual Scheimpflug analyzer, and Orbscan II for central and peripheral (Galilei vs. Orbscan) corneal thickness (CCT and PCT) measurement in normal and keratoconic eyes.

Methods:

In this prospective study, CCT and PCT were measured in 88 eyes of 88 refractive surgery candidates and 128 eyes of 69 keratoconic patients with ultrasonic pachymetry, the Galilei, and Orbscan II. The readings by the three instruments were compared using one-way analysis of normal variance. Agreement among the three devices was assessed using Pearson and intraclass correlation coefficients, and Bland–Altman plots. The same analyses were employed to evaluate agreement between Galilei and Orbscan II for PCT measurement.

Results:

In the normal group, mean CCT was 551.0±39.4, 566.9±33.5, and 565.5±40.9 μm measured by ultrasonic pachymetry, the Galilei, and Orbscan II, respectively (P<0.001). The corresponding figures in the keratoconus group were 492.0±61.7, 502.0±42.1, and 470.6±56.9 μm, respectively (P<0.001). Mean PCT was 612.5±35.3 and 640.9±38.0 μm in the normal group (P<0.001) and 567.6±35.2 and 595.1±41.4 μm in the keratoconus group (P<0.001) by the Galilei and Orbscan II, respectively. CCT and PCT measurements obtained by different devices were significantly correlated in both groups.

Conclusion:

To measure CCT, the Galilei and Orbscan II can be used interchangeably in normal eyes, but not in keratoconic eyes. For PCT, there is a systematic error between measurements obtained by the Galilei and Orbscan II. However, it is possible to change optical pachymeter readings into those obtained by ultrasonic pachymetry using a constant.

Correlation between Corneal Topographic Indices and Higher-Order Aberrations in Keratoconus

PURPOSE

To compare corneal higher-order aberrations (HOAs) between normal and keratoconic eyes, and to investigate the association between elevation-based corneal topographic indices and corneal wavefront data in the latter group.

METHODS

In this cross-sectional comparative study, 77 normal right eyes of 77 control subjects and 66 eyes of 36 keratoconic patients were included. In each eye, elevation- based corneal topographic indices including mean keratometry readings, best-fit sphere, maximum elevation, and 3-mm and 5-mm zone irregularity indices were measured using Orbscan II. The Galilei Scheimpflug analyzer was used to measure HOAs of the corneal surface. The independent student t-test was used to compare HOAs between the study groups. Spearman correlation was used to investigate possible associations between Orbscan and Galilei data in the keratoconus group.

RESULTS

All Zernike coefficients up to the 4th order except for horizontal trefoil, and vertical and horizontal tetrafoil were significantly greater in the keratoconus group than normal eyes (P<0.05). Root mean square (RMS) of HOAs up to the 6th order and total HOAs were significantly higher in the keratoconus group (P<0.05). In the keratoconus group, the strongest association was observed between vertical coma (r=-0.71, P<0.01) and total RMS of HOAs (r=0.94, P<0.01) with irregularity in the 3-mm zone. Spherical and vertical coma aberrations were significantly correlated with mean keratometry (P<0.05 for both comparisons).

CONCLUSION

Centrally located corneal HOAs are significantly greater in keratoconic eyes than normal controls. Anterior and inferior displacement of the cornea causes the majority of higher-order aberrations observed in keratoconus.