Comparison of composite and segmental methods for acquiring optical axial length with swept-source optical coherence tomography

This study compared the optical axial length (AL) obtained by composite and segmental methods using swept-source optical coherence tomography (SS-OCT) devices, and demonstrated its effects on the post-operative refractive errors (RE) one month after cataract surgery. Conventional AL measured with the composite method used the mean refractive index. The segmented-AL method used individual refractive indices for each ocular medium. The composite AL (24.52 ± 2.03 mm) was significantly longer (P < 0.001) than the segmented AL (24.49 ± 1.97 mm) among a total of 374 eyes of 374 patients. Bland–Altman analysis revealed a negative proportional bias for the differences between composite and segmented ALs. Although there was no significant difference in the RE obtained by the composite and segmental methods (0.42 ± 0.38 D vs 0.41 ± 0.36 D, respectively, P = 0.35), subgroup analysis of extremely long eyes implanted with a low power intraocular lens indicated that predicted RE was significantly smaller with the segmental method (0.45 ± 0.86 D) than that with the composite method (0.80 ± 0.86 D, P < 0.001). Segmented AL with SS-OCT is more accurate than composite AL in eyes with extremely long AL and can improve post-operative hyperopic shifts in such eyes.


Results
A total of 374 eyes of 374 patients (251 women [67.1%]) were included in this study. Patients had a mean age of 76.1 ± 8.3 years (range: 40-96 years). All eyes had successful phacoemulsification with IOL implantation. AcrySof Toric and AN6MA IOLs were implanted in 358 and 16 eyes, respectively. Table 1 summarises the ocular dimensions including the corneal thickness, ACD, AQD, lens thickness, and vitreous length measured with SS-OCT and AS-OCT. These values were significantly different when measured with the two instruments (P < 0.001). The intraclass correlation coefficient was 0.992 for the corneal thickness, 0.999 for AQD, and 0.998 for the lens thickness, indicating good agreement for these parameters.
The mean composite AL RPE was 24.81 ± 1.94 mm (range: 21.52 to 32.93 mm) and the mean segmented AL RPE was 24.79 ± 1.97 mm (range: 21.42 to 32.99 mm). These values were significantly different (P < 0.001), but there was a significant correlation between the two values (R = 1.00). The mean composite AL ILM was 24.52 ± 2.03 mm (range: 21.09 to 33.01 mm) and the mean segmented AL ILM was 24.49 ± 1.97 mm (range: 21.12 to 32.69 mm). These values were also significantly correlated (R = 1.00, P < 0.001). The mean difference between the composite and segmented AL ILM was 0.038 ± 0.067 mm.
The Bland-Altman analysis revealed a negative proportional bias between the segmented and composite AL, indicating that the difference between segmented and composite AL ILM measurements increased as AL increased (Fig. 1). The segmented AL ILM was up to 0.32 mm shorter in an eye with a composite AL ILM of 33.01 mm which is the longest eye in this study.
The mean absolute refractive error (MAE) was 0.42 ± 0.38 D using the composite AL ILM , and 0.41 ± 0.36 D using the segmented AL ILM . There was no significant difference in the MAE (P = 0.35).
A subgroup analysis was performed for the 15 extra-long eyes implanted with low power IOLs (≤+4.00 D). The same model of IOL (AN6MA) were implanted in these 15 eyes. The mean composite AL ILM was 31.02 ± 0.85 mm (range: 29.80 to 33.01 mm) and the mean segmented AL ILM was 30.78 ± 0.82 mm (range: 29.61 to 32.69 mm, Table 2), which was significantly different (P < 0.001). The mean difference between composite and segmented AL ILM measurements in extra-long eyes was 0.24 ± 0.03 mm.
The MAE using the segmented AL ILM was 0.45 ± 0.86 D, which was significantly smaller than the MAE using the composite AL ILM (0.80 ± 0.86 D, P < 0.001). Additionally, the percentage of correct refraction predictions within ±0.25 D, ±0.50 D, and ±1.00 D were 53.3%, 66.7%, and 86.7%, respectively, with segmented AL ILM measurements and 33.3%, 53.3%, and 73.3%, respectively, with composite AL ILM measurements (Table 2). There were significant differences in the percentage of refractive prediction errors between segmented and composite measurements within ±0.50 D (P = 0.003) and ±1.00 D (P = 0.01), but not ±0.25 D (P = 0.18).  Table 1. Comparison of optical biometry measurements acquired with a swept-source optical coherence tomography-based biometer and swept-source anterior segment optical coherence tomography (n = 374 eyes). Data are presented as the mean standard deviation as applicable. SS-OCT = swept-source optical coherence tomography, AS-OCT = anterior segment optical coherence tomography, Min = minimum, Max = maximum, R = correlation coefficient.

Discussion
The AL can be calculated in a variety of ways, as shown in Fig. 2. These differences arose from changes in technology and measurement compatibility with IOL power calculation formulas. Since the segmental method with the ultrasound immersion technique is more accurate than the composite method with the ultrasound contact technique, we speculated that the optical measurements with the segmental method would be more precise than the conventional ones with the composite method. Therefore, the present study compared the composite and segmental methods of obtaining optical AL ILM measurements. We found that the difference between the segmental and composite methods in optical AL ILM measurements had a negative proportional bias with AL. This is because the regression Eq. (1) was used to adjust the composite AL RPE to the composite AL ILM . As the AL RPE coefficient is 1.0446 (calculated by dividing 1.0 by 0.9573), AL ILM was overestimated in the longer eye using Eq. (1). On the other hand, regarding vitreous length, which is often longer in the longer eye (see Supplementary Fig. S1), the calculated segmented AL is longer than the composite AL due to the difference between the reflective index of the vitreous (1.336), and the mean refractive index of the whole eye (1.349). However, this difference is smaller than the effect of the regression Eq. (1). Therefore, the AL ILM was significantly shorter when acquired with the segmental method than with the composite method in highly myopic eyes. The IOL power calculation using the Haigis formula with segmented AL ILM resulted in a smaller post-operative hyperopic shift than that using the composite AL ILM in the subgroup analysis. Many IOL power calculation formulas were developed before PCI-based optical biometers were invented. Therefore, one needs to rely upon the ultrasound AL ILM 4 . The ultrasound AL ILM with the immersion technique is shorter than the optical AL RPE with the composite method; this difference could induce post-operative refraction errors 7 . Therefore, an established conversion algorithm between optical AL RPE measurements with the composite   Table 2. Predicted refractive outcomes using the composite and segmental methods in extremely long (axial length >29 mm) eyes implanted less than 4.0 diopter intraocular lens (n = 15 eyes). Data are presented as the mean standard deviation. The Haigis formula was used to calculate intraocular lens power. AL = axial length, CI = confidence interval, D = diopters, ILM = inner limiting membrane. (2020) 10:4474 | https://doi.org/10.1038/s41598-020-61391-7 www.nature.com/scientificreports www.nature.com/scientificreports/ method and immersion ultrasound AL ILM measurements with the segmental method has been used for a long time 7 . This algorithm, however, is subject to limitations such as an inadequate retinal thickness measurement in eyes with longer ALs. For example, an eye with a 30.0 mm AL ILM should have a retinal thickness of 49 μm (achieved composite AL RPE minus composite AL ILM ), but an eye with a 20.0 mm AL ILM should have a retinal thickness of 476 µm (10-fold difference). This may be one of the reasons why ALs in long and short eyes tend to be less accurate for predicting refractive outcomes following cataract surgery. In previous studies, the segmented ALs measured by an optical biometer with an 820 nm wavelength were longer in short eyes and shorter in long eyes compared to the composite ALs 8,9 . In the current study, the relationship between segmented and composite AL ILM obtained by SS-OCT with a wavelength of 1060 nm was similar to that reported previously 8,9 .
Although the most significant source of error that contributes to post-operative refractive error comes from the AL 17 Table 3. However, the high power IOL has a big influence on the refractive outcome. Therefore, we performed a subgroup analysis on patients that received an IOL with a power less than +4.0 D.
Previous studies have documented inaccuracies of the popular IOL formulas in long eyes [19][20][21][22][23][24][25] . Although the Haigis formula generally has better refractive outcomes in long eyes, post-operative hyperopic errors proportional to the AL have been reported 23 .
The current study did not find a significant difference between predicted post-operative refractive errors when the segmental method and composite method measurements of optical AL were examined in all eyes. However, the post-operative hyperopic shift with the Haigis formula was significantly improved in highly myopic eyes when the optical AL ILM acquired with the segmental method was used. Although the simple replacement of the composite AL ILM with the segmented AL ILM in the traditional IOL power calculations would not help because many formulas are optimised for the composite AL ILM , recent studies reported that the segmented AL measured via an optical biometer with a peak wavelength of 820 nm improved refractive prediction accuracy for vergence formulas 8,9 . Our study had several limitations. First, it is a retrospective single centre study and the population in our subgroup study was relatively small. We used two different IOLs; further investigation of segmented AL measurements in a prospective multicentre study with a larger group of patients and with one model is desirable. Second, the fixation status during the measurement is an important factor for AL accuracy. The SS-OCT biometer enables us to obtain an optical B-scan image during AL measurement. Therefore, it is now possible to check the fixation status during AL measurements and to examine complete longitudinal cross-sectional images of the eye [26][27][28][29] . Further studies that utilise these SS-OCT capabilities are needed. Third, our study only measured the effect of AL characteristics on post-operative refractive outcomes. However, other factors are known to influence the refractive outcomes in cataract surgery. The corneal power, in particular, should be investigated more carefully because prior studies have shown that standard keratometry overestimates the corneal power 24 . Finally, the mean group refractive index can vary with cataract grade 30 , and adjustments may be needed for cataract types and stages in future studies.
In conclusion, the segmented AL ILM is more accurate than the conventional one measured with the composite method when using SS-OCT, and the segmented AL measurement reduces the post-operative hyperopic shift in eyes with extra-long AL.

Methods
Study participants. This was a retrospective, consecutive case series of all patients who had undergone uncomplicated cataract surgery at a single centre, the National Hospital Organization, Tokyo Medical Center between October 2015 and January 2018. The protocol was reviewed and approved by the Institutional Review Board of the National Hospital Organization, Tokyo Medical Center, and was designed in accordance with the tenets of the Declaration of Helsinki. Informed consent was obtained from all patients. The selection criteria of this study followed the recommendations of a recent editorial by Hoffer et al. 31 regarding best practices for studies of IOL formulas: the use of optical biometry and the inclusion of only 1 eye from each study subject. If patients underwent bilateral cataract surgery, then a randomly selected eye was chosen for inclusion in the study. The exclusion criteria were a best-corrected distance visual acuity after cataract surgery worse than 20/40, a history of ocular surgery, a history of ocular trauma, the presence of a significant ocular comorbidity, unreliable or undetectable preoperative biometry measurements, or a history of intra-or post-operative complications.
All patients underwent cataract surgery through a 2.2-mm corneal incision. One of the following IOLs was implanted: AcrySof Toric (SN6A T3-T6, Alcon, Fort Worth, TX) or AN6MA (KOWA, Nagoya, Japan). All surgical procedures were performed under topical anaesthesia by the same experienced surgeon (TN) and all IOLs were successfully inserted into the capsular bag after phacoemulsification.  Table 3. Theoretical change in refractive outcome with a 1-mm anterior shift in intraocular lens position. Refractive outcomes are presented as the spherical equivalent. Theoretical refractive outcomes were calculated using optical design software with the ray-tracing method. All calculations were made assuming an axial length of 30 mm, an anterior corneal curvature radius of 7.8 mm, and a posterior corneal curvature radius of 6.5 mm. D = diopters.
Swept-source optical coherence tomography based biometer. The SS-OCT based biometer was used to measure AL and corneal power, along with corneal thickness, ACD, and lens thickness using the swept-source laser (1060 nm wavelength). The biometer obtained 10 consecutive scans and automatically calculated the average value.
Anterior segment optical coherence tomography. The angle analysis mode in CASIA2 was used to obtain anterior segment images comprising 16 consecutive meridional scans. This instrument uses a super-luminescent diode light source (1310 nm wavelength) and has a scan speed of 50,000 A-scans/second. All OCT images were obtained while the pupil was dilated (topical 0.5% tropicamide and 0.5% phenylephrine hydrochloride). All eyes were imaged in a dark room (illuminance of 0.3 lx) using internal fixation and were obtained twice by a trained technician who was masked to the clinical data.
Only images taken with a horizontal (180°) alignment were used in the analysis. Images were centred on the corneal vertex, which was defined as the cross point of the vertex normal and anterior corneal surface. The AS-OCT parameters measured along the vertex normal included corneal thickness, ACD, AQD, and lens thickness. Corneal thickness was defined as the distance between the anterior and posterior corneal surfaces, ACD is the distance between the anterior corneal and anterior lens surfaces, AQD is the distance between the posterior corneal and anterior lens surfaces, and lens thickness is the distance between the anterior and posterior lens surfaces (Fig. 3). The corneal thickness, AQD, and lens thickness were measured by two independent examiners (SG, II) who were masked to the clinical data.
Definition of axial length. Figure 2 illustrates how AL was obtained using acoustic and optical measurements. To measure AL, systems used either the composite method, which utilises the mean group index, or the segmental method, which accounts for individual tissue indices of the cornea, aqueous, lens, and vitreous. Ultrasound biometry can measure AL in the following four ways: acoustic contact measurements using mean group sound velocity ( Fig. 2A), acoustic immersion measurements using mean group sound velocity (Fig. 2B), acoustic segmented contact measurements using individual sound velocities (Fig. 2C), and acoustic segmented immersion measurements using individual sound velocities (Fig. 2D). Segmented immersion is theoretically the most accurate acoustic measurement because the ultrasound probe does not directly contact the cornea (Fig. 2D).
On the other hand, the SS-OCT based optical biometer measures the optical path length of the whole eye. Currently, it is essential for the optical AL measurement to determine AL ILM from the AL RPE , as popular IOL power formulas do not use AL RPE but use AL ILM measurements instead. To estimate composite distance measurements between the cornea and RPE (AL RPE , Fig. 2E), the SS-OCT based biometer system uses a mean group refractive index of 1.3496. Then, the following equation (Eq. 1), which is publicly available through the manufacturer, was used to translate AL RPE into composite distance measurements between the cornea and ILM (AL ILM, Fig. 2F) in the current study: The segmented AL RPE is measured as the sum of the thickness of the cornea, aqueous, lens, and vitreous (Fig. 2G). The refractive indices of the cornea, aqueous, lens, and vitreous were set to 1.376, 1.336, 1.410, and 1.336, respectively 26,32 . Refractive indices of the individual tissue were used during the calculation. The corneal thickness, AQD, and lens thickness were directly measured using AS-OCT. The optical vitreous length was calculated by subtracting the optical path length of the corneal thickness, AQD, and the lens thickness obtained by AS-OCT from the optical path length of the whole eye obtained using an SS-OCT based biometer. Finally, the optical vitreous length was divided by the refractive index of vitreous (1.336) to obtain the vitreous length. The segmented AL ILM (Fig. 2H) was then calculated by subtracting the retinal thickness (previously established as 300 μm [33][34][35] ) from the segmented AL RPE . formula calculations. The Haigis formula 7 was calculated using Excel spreadsheets (Microsoft Corporation, Redmond, WA, USA), and was checked against licensed commercial software on the SS-OCT based biometer. The optimised lens constants from the User Group for Laser Interference Biometry (ULIB) were used for calculations (AcrySof Toric: a0 = 1.78, a1 = 0.40, and a2 = 0.10 and AN6MA: a0 = 1.57, a1 = 0.40, and a2 = 0.10) 36 .

Data analysis.
Composite and segmented AL ILM measurements were compared by using the paired t-test.
Bland-Altman plots were used to assess the agreement between the composite and segmented AL ILM . The mean post-operative refractive errors (post-operative spherical equivalent minus the predicted post-operative spherical equivalent), mean absolute prediction error (MAE), median absolute prediction error, and standard deviation of the prediction error were calculated with the composite AL ILM and the segmented AL ILM . The Wilcoxon test was used to compare the MAE values . Subgroup analysis was performed on eyes in which low power IOL less than +4.0 D were implanted. The percentages of eyes with arithmetic prediction errors within ±0.25 D, ±0.50 D, and ±1.0 D were compared between the composite AL ILM and the segmented AL ILM using the McNemar test. Statistical significance was defined as P < 0.05. All statistical analyses were performed using JMP Pro statistical software (version 10.0.2, SAS Institute Inc., Cary, NC, USA).

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.