Artificial Intelligence as a Tool to Study the 3D Skeletal Architecture in Newly Settled Coral Recruits: Insights into the Effects of Ocean Acidification on Coral Biomineralization
Round 1
Reviewer 1 Report
Recommendation: Accept after minor revision
Comments: This manuscript describes the application of synchrotron phase contrast-enhanced microCT (PCE-CT) in combination with artificial intelligence (AI) to describe the Rapid Accretion Deposits (RADs) and Thickening Deposits (TDs) in the stony coral Stylophora pistillata. The authors demonstrate nicely how PCE-CT is able to distinguish the two main region of skeleton growth by contrast differences that are visible but not segmentable by conventional gray-level thresholding. By using deep-learning neural networks for segmentation highly-detailed, quantitative 3D information can be obtained that show the differences in RAD and TD architecture under natural and ocean acidification conditions.
The characterization results presented by the authors are high-quality and discussed in great detail. The language used within the article is clear and concise. The conclusions reached by the authors are reasonable. I believe this to be a superb manuscript and a well-written analysis of 3D PCE-CT datasets giving insight into the effects of environmental change on coral skeleton development. It is also a contribution for current and future PCE-CT users and other 3D imaging techniques in regards of strategies for data acquisition and analysis. For this reason, in my opinion the work is well-suited for JSME.
This manuscript is therefore recommended for publication in JSME with a few minor suggestions for improvement:
Figure 8: The discussion of Figure 8 was the hardest one to grasp, maybe the authors could go into more detail here, especially regarding the statistical tests they performed to analyze the estimation of the area ratios.
Additional minor remarks:
Page 1, line 20: abbreviation OA not introduced, please use ocean acidification instead
Page 1, line 21: recruits à recruit’s
Page 1, line 26: RADs and TDs à RAD’s and TD’s
Page 1, line 36f.: , are known to be among the largest bioconstructions in the world
Page 2, line 46: stable amorphous calcium carbonates – why plural? Which different ACCs exist?
Page 2, line 47: calcum à calcium
Page 5, line 179: 1s à 1 s
Page 5, line 182: enhnacement à enhancement
Page 7, line 206: dark silhouette
Page 8, line 253: please introduce the definition of ground truth images here, it appears in the figure description of Fig. 7 but would be beneficial here.
Page 12, line 340: left images
Page 13, line 366: Although
Page 13, line 370: distribution
Page 15, line 389: separation and quantification
Page 16, line 415: no linebreak before unit (100 µm)
Page 19, line 504: up to now
Page 19, line 515: , e.g.,
SI, Fig. S3: recruit’s … and , i.e., the …
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
This is a nice paper showing how artificial intelligence can assist in the segmentation of tomograms from calcified tissues. As volumetric imaging approaches have been gaining traction through different techniques – uCT, FIB-SEM, TEM tomography – approaches to improve the quality and accuracy of the segmentation are becoming increasingly important. It is nice to see this done on coral skeletons. I find the paper too technical, though, and not easy to read. Perhaps this cannot be avoided but it would be good if the authors could improve on this. More important, however, is that the paper does not deliver all that it promises. The main focus is on the segmentation of the tomograms, which is fine. But the second aim of the paper is to compare corals grown under different conditions – natural and those mimicking ocean acidification. This second part of the work remained buried in the middle of the rest, it is difficult to understand what the main conclusions really are when comparing the skeletons from both species. Indeed, most of the discussion is spent on the approach and there are only a few lines on how ocean acidification conditions impacted the skeleton growth and structure, and even then the message is not clear. I am also not entirely convinced that the number of samples analysed is enough to draw any conclusions, taking into account the biological variability. In my view, the authors need to either completely remove the part on the ocean acidification and make the deep learning segmentation approach the sole focus of the paper, or they need to expand and clarify the sections on comparing the species grown under different conditions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper describes the application of the artificial intelligence for revealing the 3D internal skeletal architecture in newly settled coral recruits. The outline of this paper is the bioimaging insights into the effect of Ocean Acidification on coral biomineralization.
I would recommend to the authors the inclusion of the figures S2 and S3 in the paper. Also, I recommend a deep conclusion of the applicability of this study on coral biomineralization.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
After reading the revised version of the paper, I still feel that the message on the effect of ocean acidification on the coral skeletons is a bit lost, but I don’t see this as the main point of the paper so I leave it to the authors to decide if they want to improve on this or not. On the second point that I raised about the number of specimens analysed, I disagree with the authors. To identify if the low pH had an effect on the structure of the corals, a significant number of specimens needs to be analysed – not 1000 tomographic slices, which are still from one specimen! That is not how statistical analysis on populations are done! What the authors need to do is tone down their statements when they say that there is a “significant morphologically-altering effect of OA on coral skeletal features” – taking into account biological variability and with n=3 per pH treatment, the results are no more than an indication that needs to be confirmed by studying a larger number of specimens.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
