Restoring drifted electron microscope volumes using synaptic vesicles at sub-pixel accuracy

Imaging ultrastructures in cells using Focused Ion Beam Scanning Electron Microscope (FIB-SEM) yields section-by-section images at nano-resolution. Unfortunately, we observe that FIB-SEM often introduces sub-pixel drifts between sections, in the order of 2.5 nm. The accumulation of these drifts significantly skews distance measures and geometric structures, which standard image registration techniques fail to correct. We demonstrate that registration techniques based on mutual information and sum-of-squared-distances significantly underestimate the drift since they are agnostic to image content. For neuronal data at nano-resolution, we discovered that vesicles serve as a statistically simple geometric structure, making them well-suited for estimating the drift with sub-pixel accuracy. Here, we develop a statistical model of vesicle shapes for drift correction, demonstrate its superiority, and provide a self-contained freely available application for estimating and correcting drifted datasets with vesicles.

The proper reconstruction of 3D electron microscope image is surely an important processing step, and therefore the paper of interest to many researchers. The text is well organized and easy to comprehend. However, it suffers from various omissions and understatements that need to be clarified, before it merits for publication in Communications Biology.
Specific comments: -Line 8-Abstract: 'in the order of 0.5 pixels'-authors should provide the physical (in nm) order of the drift, not in pixels. If we perform the scanning with a different resolution-e.g. the pixels size is 10 nm instead of 5 nm, will then the drift increase by the factor of 2? By providing the drift in pixels, the authors, improperly, suggest that the drift is related to the scanning resolution. Moreover (in the Introduction), the authors should clarify where the magnitude '0.5 pixels' comes from-their own observations, average of different values in the literature? -Line 18, 21-authors refer to FIB-SEM as a method that introduces the aforementioned drift.
-However, that drift is an imaging modality present in other serialsectioning scanning techniques, e.g. 3View.
-In the Introduction the authors should be more specific what they mean by the drift: translations, translations+dilation, etc...? They specify it in Methods, line 216, 'We define the drift in the image as a sideways translation of each image section with respect to the previous section' -this section should be moved to the Introduction.
-The authors do not provide anywhere the pixel/voxel size, nor the total sample dimensions -Figures 1-5-there are no scale bars present.
- Figure 2-The authors show the failure of the standard alignment methods, using synthetic model image. Could possibly the authors show as well such an improper alignment using real data, to support the claims made in the Introduction? -The authors provide the detailed comparison of the proposed method performance, to the performance of several standard alignment method. The authors use the simulated data, where the ground truth is known, for such a comparison. However, the comparison using real data is absent. Contrary to the case of the simulated data, the ground truth cannot be known, however, at least the qualitative comparison is available, for example an overlay of two color channel-one the aligned sections with the proposed method, the other one with the reference method (e.g. additional panels in Fig. 5: e) real data before drift correction, f) real data after drift correction with the proposed method, g) real data after drift correction with the reference method, h) pseudocolor overlay. In this way, a reader could possibly see an actual example of the proposed method superiority. This manuscript presents a restoring method for drifted FIB-SEM volume image with assumption of vesicle shapes.
Frankly, this method is only effective for FIB-SEM volume images where only spherical vesicles were observed. To extend general volume images that include other bio-components with spherical vesicles, I considered that it was necessary to include a method for automatically determining (segmenting) the vesicles. According to the Method section, the authors annotated vesicle boundary. However, detailed results for Figure 5 were not presented. Therefore, vesicles that have changed shape are indistinguishable from other vesicles. For these, I think we need to add results and discussions.
In addition, it is desirable to add samples for different biological observations in order to publish in a journal for biology.
I believe that this manuscript should be revised majorly.
The comments were given here.
1. I considered that the title of this manuscript was too exaggerated. At the very least, it should be clear that this manuscript was restoring with assuming the shape of a vesicle. For example, "with assumption of vesicle shapes" cab be inserted in the title.

Reviewer 2 Comments
Overview This manuscript presents a restoring method for drifted FIB-SEM volume image with assumption of vesicle shapes.

Comment 2.1
Frankly, this method is only effective for FIB-SEM volume images where only spherical vesicles were observed.
Response: It is correct that in this paper, we consider spherical vesicles, that is, vesicles which on average are spherical. It is certainly an interesting project to investigate other simple geometries for sub-pixel accuracy in the registration method. Initial investigations have shown that, assuming random rotations, similar properties are obtained for less symmetric objects, however, it is beyond the scope of this work to generalize to other shapes.

Comment 2.2
To extend general volume images that include other bio-components with spherical vesicles, I considered that it was necessary to include a method for automatically determining (segmenting) the vesicles.
Response: With this paper we wish to focus on the limitation of state-of-the-art registration methods, which is, that they are agnostic to the meaning of the image content. For this, an automatic method is not needed.

Comment 2.3
According to the Method section, the authors annotated vesicle boundary. However, detailed results for Figure 5 were not presented. Therefore, vesicles that have changed shape are indistinguishable from other vesicles. For these, I think we need to add results and discussions.
Response: Perhaps we have been unclear. We are performing a global transformation on each section. We annotate some vesicles, but all vesicles change shape, since we translate each image section accordingly. Therefore, in Figure 5, all vesicles have changed shape such that the average shape near a given section is spherical. Vesicles used for correction is not necessarily visible in the shown image. To illustrate the effect of the global transformation, we have updated Figure 2 to demonstrate the global effect on a real example, and we have clarified Figure 5 with added panels.

Comment 2.4
In addition, it is desirable to add samples for different biological observations in order to publish in a journal for biology.
Response: Our focus is on a common methodology used in the biology domain: registration. A limit of state-of-the-art registration methods is that they do not model what is being imaged. We explain and demonstrate this limitation, and our paper demonstrates a method to improve the results. Although we do not examine a particular biological problem, we find it an important result particularly for the audience of biology journals. However, drift is observed in other modalities, and in line 19, we have included an extended list.

Comment 2.5
I believe that this manuscript should be revised majorly.
Response: Thank you for your comments, we have considered every point carefully and made appropriate improvements.

Comment 2.6
I considered that the title of this manuscript was too exaggerated. At the very least, it should be clear that this manuscript was restoring with assuming the shape of a vesicle. For example, "with assumption of vesicle shapes" cab be inserted in the title.
Response: We agree that the title can be more specific, since we rely on statistical and size properties of objects, and thus, vesicles are a prime target. We have therefore changed the title to "Restoring drifted electron microscope volumes using synaptic vesicles at sub-pixel accuracy"

Comment 2.7
Regarding Figure 5, annotated vesicles should be marked to distinguish the other vesicles.
Response: Thank you, since the vesicles are randomly selected by our algorithm, there is no annotated vesicle in a given restored subvolume of the size shown in Figure 5. However, the effect of the annotated vesicles determines the estimated drift throughout the entire section. As discussed in response to Comment 2.8, we have on average annotated about 8 vesicles per section in the real data set. The improved visualization of this effect is given in Figure 5e-f.

Comment 2.8
The authors should discuss about performance-relations between drift-restoring and vesicle-detections.
Response: Thank you. We have already reported the total time for annotating 451 vesicles (line 119) and the relation between the number of vesicles and accuracy of our drift correction method (Supplementary Figure 8). We have added a comment on our approximate annotation speed (line 119) and a more precise discussion on the relation between the error and the number of vesicles annotated, see caption for Supplementary Figure 8 and accompanying text lines 143-149.

Comment 2.9
It is preferable to present restored images (corresponding to Suppl. -Fig. 3) by several methods in SI.
Response: We agree that it is important to visually confirm the difference in effect of standard methods with our methods. Thus, we have included new panels, see Figure 5e-h.