Corneal Endothelial Microscopy: Does a Manual Recognition of the Endothelial Cells Help the Morphometric Analysis Compared to a Fully Automatic Approach?

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Corneal endothelial microscopy: does a manual recognition of the endothelial cells help the morphometric analysis compared 3 to a fully automatic approach? 

Lines 32 and 36, the references #1 and #2  were published more than 40 and 30 years ago. Consider citing newer studies like those shown below:

Rates ERD, Almeida CD, Costa EPF, Farias RJM, Santos-Oliveira R, Alencar LMR. Layer-by-Layer Investigation of Ultrastructures and Biomechanics of Human Cornea. Int J Mol Sci. 2022 Jul 15;23(14):7833. doi: 10.3390/ijms23147833. PMID: 35887181; PMCID: PMC9317547.

Jeang LJ, Margo CE, Espana EM. Diseases of the corneal endothelium. Exp Eye Res. 2021 Apr;205:108495. doi: 10.1016/j.exer.2021.108495. Epub 2021 Feb 14. PMID: 33596440; PMCID: PMC8044020.

Petrela RB, Patel SP. The soil and the seed: The relationship between Descemet's membrane and the corneal endothelium. Exp Eye Res. 2023 Feb;227:109376. doi: 10.1016/j.exer.2022.109376. Epub 2022 Dec 30. PMID: 36592681; PMCID: PMC10287024.

Vaiciuliene R, Rylskyte N, Baguzyte G, Jasinskas V. Risk factors for fluctuations in corneal endothelial cell density (Review). Exp Ther Med. 2022 Feb;23(2):129. doi: 10.3892/etm.2021.11052. Epub 2021 Dec 10. PMID: 34970352; PMCID: PMC8713183.

Bourne WM. Biology of the corneal endothelium in health and disease. Eye (Lond). 2003 Nov;17(8):912-8. doi: 10.1038/sj.eye.6700559. PMID: 14631396.

 

Line 57. It reads: In clinical practice, an in vivo morphometric assessment and characterization of the endothelium can be carried out by specular microscopy [2,29–31]. This technique, which can be performed either with a slit lamp or with stand-alone devices such as contact and non-contact specular microscopes, allows for the observation of the corneal endothelium using the principle of specular reflection[32]. Nowadays, non-contact specular microscopes are widely used due to their non-invasive nature—eliminating the need for anaesthesia and minimizing the risks of corneal trauma or cross-infection These microscopes feature automatic image focusing technology that requires minimal practical skills, provide a wide field of view, and offer automated cell detection along with computer-assisted calculation of key morphological parameters. Many of those, such as the Konan Noncon Robo SP8000, CC7000 and CellChek XL, the Topcon Topcon SP-1000, SP 2000P Image-NET and SP3000P, the Tomey EM-3000, the CSO Perseus, and the NIDEK CEM-530, have been studied to assess measurement reliability and interchangeability[33–40]. One of the critical points of these instruments is represented by the identification of cell boundaries, which is made automatically and represents the basis of morphometric analysis. These devices also have the possibility to manually edit the boundaries of identification increasing the number of cells available for the analysis, though the criteria to identify boundaries by the manual and automatic systems might be different. In the past, some authors found out that some fully automated devices might be unreliable, and semiautomated or manual methods, though time-consuming, are more accurate [41,42].

COMMENT: It is important to highlight that specular reflection at the slit lamp, a technique introduced by Vogt over a century ago, allows for the subjective evaluation of corneal endothelial cells, but the advent of automated devices, which can assess multiple metrics of these cells, marked a significant advance in their clinical in vivo examination. In addition, it is crucial for the reader to emphasize that  analyzing specular microscopy images can be particularly challenging when guttae are present.

Consider modifying to: “In clinical practice, an in vivo morphometric subjective assessment and characterization of the corneal endothelium can be performed using specular reflection at the slit lamp. This technique has been utilized since Vogt’s original description over a century ago (Vogt, 1920; Sugar, 1979) [2, 29-31]. The introduction of standalone devices for quantitative analysis, initially by Maurice in 1968 (Maurice 1968), allowed the examination of enucleated eyes, and soon after, these were adapted for both contact and non-contact use in patients (Laing et al., 1975; Olsen, 1979). Over the past four decades, non-contact specular microscopes have gained popularity in clinical settings due to their non-invasive nature, eliminating the need for topical anesthesia and the risk of epithelial trauma and cross-infection. These microscopes, which include models like the Konan Noncon Robo SP8000, CC7000, CellChek XL, Topcon SP-1000, SP 2000P Image-NET, SP3000P, Tomey EM-3000, CSO Perseus, and NIDEK CEM-530, feature automatic image focusing that requires minimal technical skill, provides a wide field of view, and offers automated cell detection with computer-assisted calculations of key morphological parameters. Studies have been conducted on these models to assess their measurement reliability and interchangeability [33-40]. A critical aspect of these devices is their ability to identify cell boundaries automatically, which forms the basis for morphometric analysis. While these devices offer manual editing of cell boundary identification, increasing the number of analyzable cells, the criteria for boundary identification may differ between manual and automatic systems. Historically, some fully automated systems have been deemed unreliable, with semi-automated or manual methods, though more time-consuming, providing greater accuracy [41, 42]. Furthermore, analyzing the endothelium in the presence of guttae is even more complex, as the software may not recognize guttae as abnormal areas devoid of normal endothelial cells, necessitating manual identification (McLaren et al., 2014; Fujimoto et al., 2014) or the application of recently developed artificial intelligence-based approaches (Prada et al., 2024).

Additional references to be cited:

Sugar A. Clinical specular microscopy. Surv Ophthalmol. 1979 Jul-Aug;24(1):21-32. doi: 10.1016/0039-6257(79)90144-9. PMID: 384572.

Vogt, A. Die Sichtbarkeit des lebenden Hornhautendothels. Graefes Arhiv für Ophthalmologie 1920;101, 123–144.

 

Maurice D: Cellular membrane activity in the cornea1 endothelium of the intact eye. Experientia 1968; 24: 1094- 1095.

Laing RA, Sandstrom MM, Leibowitz HM: In vivo photomicrography of the cornea1 endothelium. Arch Ophthalmol 1975;93:143-145.

Olsen T. Non-contact specular microscopy of human corneal endothelium. Acta Ophthalmol (Copenh). 1979;57(6):986-98. doi: 10.1111/j.1755-3768.1979.tb00529.x. PMID: 546011.

McLaren JW, Bachman LA, Kane KM, et al. Objective assessment of the corneal endothelium in Fuchs’ endothelial dystrophy. Invest Ophthalmol Vis Sci. 2014;55:1184–1190.

Fujimoto H, Maeda N, Soma T, et al. Quantitative regional differences in corneal endothelial abnormalities in the central and peripheral zones in Fuchs’ endothelial corneal dystrophy. Invest Ophthalmol Vis Sci. 2014;55:5090–5098.

Prada AM, Quintero F, Mendoza K, Galvis V, Tello A, Romero LA, Marrugo AG. Assessing Fuchs Corneal Endothelial Dystrophy Using Artificial Intelligence-Derived Morphometric Parameters From Specular Microscopy Images. Cornea. 2024 Sep 1;43(9):1080-1087.

Line 352. It reads: “A potential limitation of this study is that the first sample studied is made by subjects with normal corneas and the images (in all the corneal portions) were characterized by high quality. For instance, the number of cells automatically detected was ≥75 units, a threshold above which parameter estimates become highly reliable[44]. It is acknowledged that automated methods may encounter challenges in determining endothelial parameters from poor-quality images[39]. The present study did not examine a sample with poor quality images, and even the sub-sample of images with guttae comprised only slightly degraded images, with a maximum of 5.0% of the total area covered by guttae. A further limitation is that only one operator carried out the study, so no information about inter-operator variability is available.”

 

COMMENT: It is important to indicate that the approach used in the endothelium with guttae, has been shown to be inaccurate, since excluding the area covered by guttae from the calculation will cause an overestimation of cell density.

Therefore, that weakness should be recognized, and the text should read:

"A potential limitation of this study is that the initial sample consisted of subjects with normal corneas, and the images from all corneal regions were of high quality. For example, the number of cells automatically detected was ≥75 units, which is above the threshold needed for reliable parameter estimation [44]. It is known that automated methods may struggle to accurately determine endothelial parameters from poor-quality images [39]. This study did not include a sample of poor-quality images, and even the subset of images with guttae included only slightly degraded images, with a maximum of 5.0% of the total area affected by guttae.

Another limitation is that only a single operator conducted the study, so no data on inter-operator variability is available. Additionally, in the subgroup of specular images with guttae, the approach of excluding areas covered by guttae from calculations is acknowledged as a limitation. This method can lead to overestimation of cell density, as has been shown in previous research (Prada et al., 2024)."

 

Additional reference to be cited:

Prada AM, Quintero F, Mendoza K, Galvis V, Tello A, Romero LA, Marrugo AG. Assessing Fuchs Corneal Endothelial Dystrophy Using Artificial Intelligence-Derived Morphometric Parameters From Specular Microscopy Images. Cornea. 2024 Sep 1;43(9):1080-1087.

 

 

 

 

Author Response

We thank the Reviewer for the time spent on the manuscript.

Please find below, a point-by-point reply.
In this reply letter, Reviewer comments are indicated in black. Our answer is in light blue. In the revised
manuscript (revision 2), all changes are indicated in red (the red changes are also reported in this reply).

 

Lines 32 and 36, the references #1 and #2  were published more than 40 and 30 years ago. Consider citing newer studies like those shown below:

Rates ERD, Almeida CD, Costa EPF, Farias RJM, Santos-Oliveira R, Alencar LMR. Layer-by-Layer Investigation of Ultrastructures and Biomechanics of Human Cornea. Int J Mol Sci. 2022 Jul 15;23(14):7833. doi: 10.3390/ijms23147833. PMID: 35887181; PMCID: PMC9317547.

Jeang LJ, Margo CE, Espana EM. Diseases of the corneal endothelium. Exp Eye Res. 2021 Apr;205:108495. doi: 10.1016/j.exer.2021.108495. Epub 2021 Feb 14. PMID: 33596440; PMCID: PMC8044020.

Petrela RB, Patel SP. The soil and the seed: The relationship between Descemet's membrane and the corneal endothelium. Exp Eye Res. 2023 Feb;227:109376. doi: 10.1016/j.exer.2022.109376. Epub 2022 Dec 30. PMID: 36592681; PMCID: PMC10287024.

Vaiciuliene R, Rylskyte N, Baguzyte G, Jasinskas V. Risk factors for fluctuations in corneal endothelial cell density (Review). Exp Ther Med. 2022 Feb;23(2):129. doi: 10.3892/etm.2021.11052. Epub 2021 Dec 10. PMID: 34970352; PMCID: PMC8713183.

Bourne WM. Biology of the corneal endothelium in health and disease. Eye (Lond). 2003 Nov;17(8):912-8. doi: 10.1038/sj.eye.6700559. PMID: 14631396.

 

Response: We thank the Reviewer for this comment. We have added 2 of the suggested papers.

 

Line 57. It reads: In clinical practice, an in vivo morphometric assessment and characterization of the endothelium can be carried out by specular microscopy [2,29–31]. This technique, which can be performed either with a slit lamp or with stand-alone devices such as contact and non-contact specular microscopes, allows for the observation of the corneal endothelium using the principle of specular reflection[32]. Nowadays, non-contact specular microscopes are widely used due to their non-invasive nature—eliminating the need for anaesthesia and minimizing the risks of corneal trauma or cross-infection These microscopes feature automatic image focusing technology that requires minimal practical skills, provide a wide field of view, and offer automated cell detection along with computer-assisted calculation of key morphological parameters. Many of those, such as the Konan Noncon Robo SP8000, CC7000 and CellChek XL, the Topcon Topcon SP-1000, SP 2000P Image-NET and SP3000P, the Tomey EM-3000, the CSO Perseus, and the NIDEK CEM-530, have been studied to assess measurement reliability and interchangeability[33–40]. One of the critical points of these instruments is represented by the identification of cell boundaries, which is made automatically and represents the basis of morphometric analysis. These devices also have the possibility to manually edit the boundaries of identification increasing the number of cells available for the analysis, though the criteria to identify boundaries by the manual and automatic systems might be different. In the past, some authors found out that some fully automated devices might be unreliable, and semiautomated or manual methods, though time-consuming, are more accurate [41,42].

COMMENT: It is important to highlight that specular reflection at the slit lamp, a technique introduced by Vogt over a century ago, allows for the subjective evaluation of corneal endothelial cells, but the advent of automated devices, which can assess multiple metrics of these cells, marked a significant advance in their clinical in vivo examination. In addition, it is crucial for the reader to emphasize that  analyzing specular microscopy images can be particularly challenging when guttae are present.

Consider modifying to:

“In clinical practice, an in vivo morphometric subjective assessment and characterization of the corneal endothelium can be performed using specular reflection at the slit lamp. This technique has been utilized since Vogt’s original description over a century ago (Vogt, 1920; Sugar, 1979) [2, 29-31]. The introduction of standalone devices for quantitative analysis, initially by Maurice in 1968 (Maurice 1968), allowed the examination of enucleated eyes, and soon after, these were adapted for both contact and non-contact use in patients (Laing et al., 1975; Olsen, 1979). Over the past four decades, non-contact specular microscopes have gained popularity in clinical settings due to their non-invasive nature, eliminating the need for topical anesthesia and the risk of epithelial trauma and cross-infection. These microscopes, which include models like the Konan Noncon Robo SP8000, CC7000, CellChek XL, Topcon SP-1000, SP 2000P Image-NET, SP3000P, Tomey EM-3000, CSO Perseus, and NIDEK CEM-530, feature automatic image focusing that requires minimal technical skill, provides a wide field of view, and offers automated cell detection with computer-assisted calculations of key morphological parameters. Studies have been conducted on these models to assess their measurement reliability and interchangeability [33-40]. A critical aspect of these devices is their ability to identify cell boundaries automatically, which forms the basis for morphometric analysis. While these devices offer manual editing of cell boundary identification, increasing the number of analyzable cells, the criteria for boundary identification may differ between manual and automatic systems. Historically, some fully automated systems have been deemed unreliable, with semi-automated or manual methods, though more time-consuming, providing greater accuracy [41, 42]. Furthermore, analyzing the endothelium in the presence of guttae is even more complex, as the software may not recognize guttae as abnormal areas devoid of normal endothelial cells, necessitating manual identification (McLaren et al., 2014; Fujimoto et al., 2014) or the application of recently developed artificial intelligence-based approaches (Prada et al., 2024).

Additional references to be cited:

Sugar A. Clinical specular microscopy. Surv Ophthalmol. 1979 Jul-Aug;24(1):21-32. doi: 10.1016/0039-6257(79)90144-9. PMID: 384572.

Vogt, A. Die Sichtbarkeit des lebenden Hornhautendothels. Graefes Arhiv für Ophthalmologie 1920;101, 123–144.

 Maurice D: Cellular membrane activity in the cornea1 endothelium of the intact eye. Experientia 1968; 24: 1094- 1095.

Laing RA, Sandstrom MM, Leibowitz HM: In vivo photomicrography of the cornea1 endothelium. Arch Ophthalmol 1975;93:143-145.

Olsen T. Non-contact specular microscopy of human corneal endothelium. Acta Ophthalmol (Copenh). 1979;57(6):986-98. doi: 10.1111/j.1755-3768.1979.tb00529.x. PMID: 546011.

McLaren JW, Bachman LA, Kane KM, et al. Objective assessment of the corneal endothelium in Fuchs’ endothelial dystrophy. Invest Ophthalmol Vis Sci. 2014;55:1184–1190.

Fujimoto H, Maeda N, Soma T, et al. Quantitative regional differences in corneal endothelial abnormalities in the central and peripheral zones in Fuchs’ endothelial corneal dystrophy. Invest Ophthalmol Vis Sci. 2014;55:5090–5098.

Prada AM, Quintero F, Mendoza K, Galvis V, Tello A, Romero LA, Marrugo AG. Assessing Fuchs Corneal Endothelial Dystrophy Using Artificial Intelligence-Derived Morphometric Parameters From Specular Microscopy Images. Cornea. 2024 Sep 1;43(9):1080-1087.

Response: We thank the Reviewer for the suggestions provided that allow to improve the introduction. We have integrated in the manuscript the sentences suggested as well as the new references.

 

In clinical practice, an in vivo morphometric assessment and characterization of the endothelium can be carried out by specular microscopy [3,31–33], which can be performed either with a slit lamp or with stand-alone devices such as contact and non-contact specular microscopes, allows for the observation of the corneal endothelium using the principle of specular reflection [34,35]. This technique has been utilized since Vogt’s original description over a century ago[32,36]. The introduction of standalone devices for quantitative analysis, initially by Maurice in 1968[37], allowed the examination of enucleated eyes, and soon after, these were adapted for both contact and non-contact use in patients[31,34]. Over the past four decades, non-contact specular microscopes have gained popularity in clinical settings due to their non-invasive nature—eliminating the need for anaesthesia and minimizing the risks of corneal trauma or cross-infection. These microscopes, which include models like the Konan Noncon Robo SP8000, CC7000, CellChek XL, Topcon SP-1000, SP 2000P Image-NET, SP3000P, Tomey EM-3000, CSO Perseus, and NIDEK CEM-530, feature automatic image focusing technology that requires minimal practical skills, provide a wide field of view, and offer automated cell detection along with computer-assisted calculation of key morphological parameters. Studies have been conducted on these models to assess measurement reliability and interchangeability[38–45]. A critical aspect of these devices is their ability to identify cell boundaries automatically, which forms the basis for morphometric analysis. While these devices offer manual editing of cell boundary identification, increasing the number of analysable cells, the criteria for boundary identification may differ between manual and automatic systems. Historically, some fully automated systems have been deemed unreliable, with semi-automated or manual methods, though more time-consuming, providing greater accuracy [46,47]. Furthermore, analyzing the endothelium in the presence of guttae is even more complex, as the software may not recognize guttae as abnormal areas devoid of normal endothelial cells, necessitating manual identification[48,49] or the application of recently developed artificial intelligence-based approaches[50].

 

 

Line 352. It reads: “A potential limitation of this study is that the first sample studied is made by subjects with normal corneas and the images (in all the corneal portions) were characterized by high quality. For instance, the number of cells automatically detected was ≥75 units, a threshold above which parameter estimates become highly reliable[44]. It is acknowledged that automated methods may encounter challenges in determining endothelial parameters from poor-quality images[39]. The present study did not examine a sample with poor quality images, and even the sub-sample of images with guttae comprised only slightly degraded images, with a maximum of 5.0% of the total area covered by guttae. A further limitation is that only one operator carried out the study, so no information about inter-operator variability is available.”

 

COMMENT: It is important to indicate that the approach used in the endothelium with guttae, has been shown to be inaccurate, since excluding the area covered by guttae from the calculation will cause an overestimation of cell density.

Therefore, that weakness should be recognized, and the text should read:

"A potential limitation of this study is that the initial sample consisted of subjects with normal corneas, and the images from all corneal regions were of high quality. For example, the number of cells automatically detected was ≥75 units, which is above the threshold needed for reliable parameter estimation [44]. It is known that automated methods may struggle to accurately determine endothelial parameters from poor-quality images [39]. This study did not include a sample of poor-quality images, and even the subset of images with guttae included only slightly degraded images, with a maximum of 5.0% of the total area affected by guttae.

Another limitation is that only a single operator conducted the study, so no data on inter-operator variability is available. Additionally, in the subgroup of specular images with guttae, the approach of excluding areas covered by guttae from calculations is acknowledged as a limitation. This method can lead to overestimation of cell density, as has been shown in previous research (Prada et al., 2024)."

Response: Again, we thank the Reviewer  for the comments on the final part of discussion concerning the limitations of the work.

We integrated this final part with all the suggestions.

 

A potential limitation of this study is that the initial sample consisted of subjects with normal corneas and the images from all the corneal regions were of high quality. For example, the number of cells automatically detected was ≥75 units, which is above the threshold needed for reliable parameter estimation [52]. It is known that automated methods may struggle to accurately determine endothelial parameters from poor-quality images[44]. This study did not include a sample of poor quality images, and even the sub-sample of images with guttae included only slightly degraded images, with a maximum of 5.0% of the total area covered by guttae. Another limitation is that only a single operator conducted the study, so no data about inter-operator variability is available. Additionally, in the subgroup of specular images with guttae, the approach of excluding areas covered by guttae from calculations is acknowledged as a limitation. This method can lead to overestimation of cell density, as has been shown in previous research[50].

 

Reviewer 2 Report

Comments and Suggestions for Authors

I have carefully reviewed the article titled "Manual versus automatic recognition of endothelial cells in corneal endothelial microscopy" and have provided detailed observations and suggestions to further improve its quality, with particular attention to enhancing the introduction, methodology, statistical analysis, and discussion sections.The introduction should emphasize more the importance of the examination in clinical practice, specifying additional applications beyond cataract surgery preparation or the detection of structural anomalies. Regarding the methods, it appears that the results were evaluated by a single trained operator. It is necessary to clarify whether all assessments were conducted by one person and what level of experience they had in interpreting the results. The authors should discuss whether there is a possibility of statistical bias due to the assessment by a single individual. These aspects should be addressed. The statistical section should be expanded with more details. In the discussion, the advantages and disadvantages should be compared in more depth. For example, does the time taken to analyze a single sample differ significantly between the two methods? I ask the authors, if unable to answer, to indicate these potential issues among the study’s limitations.

Author Response

We thank you for the time spent on the manuscript. Please find below, a point-by-point reply.

In this reply letter, Reviewer comments are indicated in black. Our answer is in light blue. In the revised manuscript, all changes are indicated in red (the red changes are also reported in this reply).

Reviewer #1:

I have carefully reviewed the article titled "Manual versus automatic recognition of endothelial cells in corneal endothelial microscopy" and have provided detailed observations and suggestions to further improve its quality, with particular attention to enhancing the introduction, methodology, statistical analysis, and discussion sections.

Response: We thank the Reviewer for his/her supportive comments.

 

The introduction should emphasize more the importance of the examination in clinical practice, specifying additional applications beyond cataract surgery preparation or the detection of structural anomalies.

Response: Thanks for providing this suggestion. In introduction, we have mentioned when corneal endothelium assessment is useful beyond cataract surgery preparation such as for ocular conditions diagnosis and in the evaluation of safety of surgical procedures.

"Corneal endothelial assessment is also of primary importance for the diagnosis and treatment of many ocular conditions such as Fuchs endothelial corneal dystrophy, Posterior Polymorphous Corneal Dystrophy, Congenital Hereditary Endothelial Dystrophy, Pseudoexfoliation Keratopathy, keratoconus or in monitoring the impact of anterior chamber inflammation on the cornea[6–9]. Furthermore, endothelial parameters are evaluated to establish the outcomes in terms of safety of other surgical procedures beyond cataract surgery such as refractive surgery, corneal collagen crosslinking, Descemet stripping automated endothelial keratoplasty (DSEK), Descemet's membrane endothelial keratoplasty (DMEK), corneal graft, and glaucoma surgery[10–19]. Changes in endothelial parameters are also investigated in response to topical drugs[20,21]

Endothelium monitoring can be a powerful tool also for contact lens wearers, since hypoxia conditions induced by contact lens wear have been linked to an increase in the variability of cell size (polymegathism) and shape (pleomorphism)[22–24] or in the monitoring of corneal structural changes after orthokeratology[25]."

 

Regarding the methods, it appears that the results were evaluated by a single trained operator. It is necessary to clarify whether all assessments were conducted by one person and what level of experience they had in interpreting the results. The authors should discuss whether there is a possibility of statistical bias due to the assessment by a single individual. These aspects should be addressed.

Response: We confirm that all the assessments were performed by one person, as already stated. However, we have added further information about the level of experience of the operator.

This operator was a pre-reg Optometrist who conducted extensive training on the instrument before starting the study, with over 80 endothelial images manually edited under the supervision of a lecturer with experience in endothelial microscopy over 25 years. This training was considered suitable to reach a good level of reliability, considering that the operator's sole task was to add and correct the identification of endothelial cells and to report the reprocessed data. The decision to use a single well-trained operator for the manual editing phase was made to avoid introducing bias that could arise from the involvement of multiple operators.”

Also, in the discussion, we reported the potential issues due to the presence of a single operator.

"A further limitation is that only one operator carried out the study, so no information about inter-operator variability is available."

 

The statistical section should be expanded with more details.

Response: Some further details have been added in the statistical section of the method.

“This was done to avoid possible bias link to the fact that measurements obtained from the right and left eye of a subject are often correlated[45].

…..

This kind of plot allow to detect any systematic trend in the differences between the fully automatic procedure and the manual procedure due to an increase in the amplitude of the outcome examined.”

 

In the discussion, the advantages and disadvantages should be compared in more depth. For example, does the time taken to analyze a single sample differ significantly between the two methods? I ask the authors, if unable to answer, to indicate these potential issues among the study’s limitations.

Response:

This point raised by the Reviewer had been already discussed in lines 284-289 of the discussion. However, to better compare advantages and disadvantages of the two procedure we have added some further lines.

 

On average the trained operator needed from 20 min to 30 min to edit, in one image, the cell boundaries missed/wrongly detected by the fully automated procedure. However, the manual editing was almost limited to the addition of cells not detected rather than the correction of wrong boundaries inserted by the automatic procedure, which occurred for a maximum of 2/3 cells for each image. In contrast, the fully automatic procedure completed the task of identify cell boundaries and calculated endothelial mosaic parameters, in a few seconds per image. Given that a trained operator required significantly more time, it can be assumed that eventually, in an untrained operator, it would take even longer time, potentially increasing both the time required and the risk of errors.

 

Reviewer 3 Report

Comments and Suggestions for Authors

Well done experiment and report on results.  Detail is exceptional.  This confirms the validity and accuracy of automated endothelial cell evaluation.  It confirms the value of automated systems in saving ophthalmologist or technologists time and effort.  Suggest adding 'endothelial' to cell density on line 17.

Author Response

We thank you for the time spent on the manuscript. Please find below, a point-by-point reply.

In this reply letter, Reviewer comments are indicated in black. Our answer is in light blue. In the revised manuscript, all changes are indicated in red.

 

Reviewer #2:

Well done experiment and report on results. Detail is exceptional. This confirms the validity and accuracy of automated endothelial cell evaluation. It confirms the value of automated systems in saving ophthalmologist or technologists time and effort. Suggest adding 'endothelial' to cell density online 17.

Response: We thank the Reviewer for his/her comments.

 

Suggest adding 'endothelial' to cell density on line 17.

Response: Done