Comparison between the CASIA SS-1000 and Pentacam in measuring corneal curvatures and corneal thickness maps

In all, 66 consecutive adult subjects (at least 20 years of age) were recruited. After obtaining informed consent, demographics, medical/ocular history, and present ocular medications were recorded. Eyes were excluded if they had any previous intraocular or corneal surgery (except laser refractive surgery), used a rigid contact lens within 1 week or soft contact lens within 24 hours, were currently on mydriatic and/or miotic medications, or were incapable of fixating at the internal fixation light or opening the eye sufficient to image a full image area. When both eyes of the participant were eligible, one eye was randomly selected by coin flip.

Study eyes underwent a series of imaging by three CASIA SS-1000 devices and three Pentacam devices. These devices were placed in three separate rooms. In each room, one CASIA SS-1000 and one Pentacam were installed and one examiner was assigned to operate the devices. Three images were taken by each of six devices. A computer-generated, randomization schedule was used to determine the order of the rooms/examiners and the order of the devices (CASIA SS-1000 vs. Pentacam) in a room. The exam procedure took approximately 1 h for imaging a total of 18 images in each eye. With this design, the device effect and examiner effect were confounded, and is called the [device+examiner] effect throughout the rest of this paper.

All measurements are performed in constant, dim-lighting conditions. Several corneal curvature and corneal thickness measurements were obtained from each image by each instrument:

Demographics were summarized by mean and standard deviation (SD) for continuous variables or by frequency (%) for discrete variables. The repeatability and reproducibility were evaluated by:

Repeatability SD and Reproducibility SD were calculated for each device using a random effect model with random effects, participants (66 participants) and devices + examiners (three devices + examiners). The ratio of CASIA SS-1000 SD to Pentacam SD for each measure was computed for repeatability, reproducibility, and agreement. A ratio of less than 1 indicates that CASIA SS-1000 is more consistent than Pentacam, and vice versa. The coefficients of variation (CVs) were also calculated by repeatability SD/mean and reproducibility SD/mean.

Agreement, mean difference (bias) and limits of agreement (LOA) between CASIA SS-1000 and Pentacam was calculated. In addition, Deming regression analysis, which allows measurement errors in both CASIA SS-1000 and Pentacam, was used to examine any systematic deviations and estimate the conversion equations. Bland-Altman agreement plots and Deming regression plots were graphed.

Statistical analysis was performed using PROC MIXED procedure in SAS for Window 9.4 (SAS Inc., Cary, NC). Bland-Altman agreement plot was generated using BA.plot() and Deming regression analysis was performed using Deming() in MethComp package using R × 64 Version 3.01.

Complete results are shown in Table 2. All repeatability SD ratios for corneal curvature and thickness measurements were less than 1, which indicates that the CASIA SS-1000 is more consistent within an eye measured by the same device and the same operator, especially the corneal thickness measures (0.22–0.30). The repeatability of anterior corneal curvature between the two instruments were similar (ratio = 0.86 and 0.85 for anterior flat and steep meridian curvature, respectively). All coefficients of variation (CV) (SD/mean) for corneal curvature and thickness measurements were less than 0.5% for CASIA SS-1000 and 1% for Pentacam. The repeatability for TCP(X) was similar between two instruments (SD ratio = 1.06 with CV = 48.8% for CASIA; 54.3% for Pentacam), and CASIA was more repeatable in TCP(Y) than Pentacam (SD ratio = 0.68 with CV = 35.6% for CASIA and 60.5% for Pentacam).

All reproducibility SD ratios were less than 1, indicating the CASIA SS-1000 is more consistent (Table 2). The ratios for corneal thickness ranged from 0.19–0.22, indicating the Pentacam’s inter-[device+examiner] variation was 5 times greater than the CASIA SS-1000. The ratios for posterior corneal curvature ranged from 0.25–0.39. All CVs for corneal thickness and curvature measurements were less than 0.5% for CASIA SS-1000 and 2.2% for Pentacam. The reproducibility of TCP(X) was similar between the two instruments (SD ratio = 0.92 and with CV = 51.2% for CASIA and 65.7% for Pentacam), while CASIA had a better reproducibility in TCP(Y) (SD ratio = 0.52 with CV = 37.8% for CASIA and 86.8% for Pentacam).

Bland-Altman agreement plots and Deming regression plots are shown in Figs. 1 and 2, respectively. Anterior corneal curvature had excellent agreement between CASIA SS-1000 and Pentacam. The mean difference between the two instruments (see Table 2 and Fig. 1) was 0.01 D (P = 0.71) and 0.02 D (P = 0.39) for flat and steep meridian, respectively, and 95% of the differences for the flat meridian were within 0.70 D and 0.87 D for the steep meridian. The CASIA SS-1000 posterior corneal curvature was flatter by 1.13 D (P < 0.001) and 1.35 D (P < 0.001) in flat and steep meridian, respectively, compared to the Pentacam. The widths of LOAs for posterior corneal curvatures were about 2.5 times wider than the LOA widths for anterior curvatures.

Corneal thickness measured by CASIA SS-1000 was thinner compared with the values obtained with the Pentacam in various locations. The mean difference in TCT between CASIA SS-1000 and Pentacam was − 11.20 μm (P < 0.001) with LOA of [− 34.9 12.5]. The mean difference in CCT between CASIA SS-1000 and Pentacam was − 10.45 μm (P < 0.001) with LOA of [− 33.75, 12.85]. The differences and LOA increased when the location of measurements was away from the central cornea (Fig. 1). The mean difference in TCP(X) and TCP(Y) between CASIA SS-1000 and Pentacam was 0.06 mm (P = 0.001) and − 0.07 mm (P = 0.004), respectively, and 95% of the differences for TCP(X) were within 0.70 mm, which was smaller than 1.07 mm for TCP(Y) (Fig. 1). Further, Bland-Altman plots showed no special distribution patterns, such as significant trend in mean difference or varying the LOA width.

Deming regression plots (Fig. 2) showed that the all slopes of Deming regression lines (blue lines) were parallel to the red lines (CASIA SS-1000 = Pentacam or slope = 1), except flat and steep posterior corneal curvatures and TCP(X). The slope of flat and steep was 0.83 and 0.87, respectively, which indicates that CASIA SS-1000 measurements are smaller than Pentacam when the posterior curvature increases. The slope of TCP(X) was 1.42 and intercepted with the line with slope = 1 around the apex (TCP(X) = 0), which implies that CASIA SS-1000 measured farther away from the apex than Pentacam. However, the difference may not be clinically significant. A linear conversion between measurements, taken by CASIA SS-1000 and Pentacam, was derived from Deming regression analysis (Table 3).

Anterior curvatures, along with axial length, are the key factors in calculating IOL power. A difference of more than 1 D in anterior corneal curvature is considered as clinically significant. The repeatability and reproducibility were excellent (SDs < 0.3 D). However, the CASIA SS-1000 had a better reproducibility in measuring flat axis (ratio = 0.58). The average difference in AKf and AKs between the two devices was 0.01 D and − 0.02 D, respectively. LOA limits are within 1 D. Thus, the two instruments are substantially equivalent, however, the CASIA SS-1000 had more consistent results.

TCT and CCT are important parameters for preoperative refractive surgery screening. In refractive surgery, based on the Munnerlyn formula [12], every 12 μm of ablation depth will correct 1 D of the spherical equivalent for a single 6 mm ablation zone. Depending on the lasers and ablation zones, the average corneal ablation depth ranged from 12 to 17 μm per diopter [12]. In addition, sufficient thickness of both flap and residual posterior stroma is required to prevent irregular astigmatism and to avoid diseases of corneal integrity and/or postoperative corneal ectasia. In general, a minimum of 110 μm flap will be created, leaving a minimum of 250 μm of residual posterior stroma. Thus, the accuracy of TCT and CCT are extremely important for thin corneas and high-myopia corrections, which require greater ablation depth. An overestimate of TCT and CCT during refractive surgery screening may result in postoperative complications. Therefore, an overestimate of 12–17 μm is considered clinically significant.

CASIA SS-1000 is 4x (repeatability ratio = 0.22–0.24) more precise than Pentacam when repeating TCT and CCT measurements using the same device, and 5x (repeatability ratio = 0.20) more precise than Pentacam when repeating TCT and CCT measurements using different devices.

The mean difference in CCT between CASIA SS-1000 and Pentacam of − 10.45 μm (~ 2% of average CCT for all participants) with LOA ranging 45 μm (− 33.75, 12.85) could be considered clinically significant. This difference may lead ophthalmologists to make a safer (more conservative) decision for refractive surgery. However, it may disqualify some patients who would be qualified if measured by Pentacam. The two measurements can be converted by the following equation: CASIA = 10.87 + 0.97 x Pentacam. For example, an eye measured 500 μm CCT by Pentacam can be converted to CASIA measurement as 490.87 μm. Therefore, decreasing the clinical difference in the values can be obtained. Similarly, TCT measurements can be converted by CASIA = 6.88 + 0.97 Pentacam. However, the decreased variability in measurements obtained with the CASIA SS-1000 results in more consistent clinical decisions.

In addition, after reviewing 94 papers published between 2000 and 2014, Rozema et al. reported that of the 14 instruments in which Pentacam was compared, five significantly overestimated CCT in normal eyes [13]. These included TMS-5 (Tomey Corp, Nagoya, Japan, by 17 μm), Orbscan with AF (Bausch & Lomb, Claremont, CA, by 11 μm), Visante/Stratus (Carl Zeiss, Dublin, CA, by 14 μm), SL-OCT (Heidelberg Engineering, Heidelberg, Germany, by 16 μm), and specular microscopes (by 20 μm). Other studies showed similar results, with Pentacam overestimating by an average of 17 μm, compared to Visante OMNI (Carl Zeiss, Dublin, CA) (17.8 ± 11.3 μm) [14], and Lenstar (Haag-Streit AG, Koeniz, Switzerland) (17.1 ± 8.5 μm) [15].

Recent studies have shown that posterior corneal curvature, especially maximal point of elevation over the best-fit reference image, was statistically significant for keratoconus [16]. The ability to detect subclinical keratoconus earlier has important implications for improving refractive surgical screening to ensure that patients do not develop subsequent ectasia [16]. In theory, flat and steep posterior corneal curvature (PKf and PKs) may provide useful information for more accurate toric IOL calculations. However, measured vs. assumed posterior corneal curvatures have shown no difference in IOL calculation accuracy [17]. Our study showed that the CASIA SS-1000 underestimated the posterior corneal curvature comparing to the Pentacam. It is difficult to know what implications that an underestimate will have on screening for early keratoconus. The topographic morphologic distribution of the posterior corneal surface, specifically looking at the superior and inferior asymmetry within one topographic image, is a better early indicator of subclinical keratoconus than the raw absolute value of aggregate posterior corneal curvature being higher. Further studies are needed to analyze the topographic map of the posterior surface—and not just a solitary absolute value of the mean corneal curvature.

The absolute value of the posterior surface being smaller will affect the IOL calculation if the posterior corneal curvature value is used to calculate the total corneal power. However, recent studies using the Barret toric calculator showed that an estimated posterior curvature is just as accurate as a measured corneal curvature when looking at post-cataract refractive outcomes. As the Barrett formula is the current standard for toric IOL calculations, the difference in absolute value measurement of the posterior corneal surface between the Pentacam and CASIA will have little clinical difference in accuracy when correcting astigmatism, as the assumption has shown equivalent results to measured values. The anterior corneal surface measurements between the Pentacam and CASIA, which are more important in corneal measurement when performing IOL calculations, were not statistically different.

The corneal thickness and location data has less variability for the CASIA, compared to the Pentacam, in both repeatability and reproducibility. The CASIA corneal thickness measurements are also consistently lower in value than the Pentacam corneal thickness measurement in every point in the periphery: PCT2, PCT4, and PCT6. The pattern/distribution of corneal thickness (pachymetry map) and the location of the thinnest point in the corneal is important in planning for refractive surgery and detecting corneal diseases, such as keratoconus [18]. Thus, less variability in both the corneal thickness measurements and location of thinnest corneal point are all very important clinical measurements for screening laser refractive surgery patients.