TEM image restoration from fast image streams

Microscopy imaging experiments generate vast amounts of data, and there is a high demand for smart acquisition and analysis methods. This is especially true for transmission electron microscopy (TEM) where terabytes of data are produced if imaging a full sample at high resolution, and analysis can take several hours. One way to tackle this issue is to collect a continuous stream of low resolution images whilst moving the sample under the microscope, and thereafter use this data to find the parts of the sample deemed most valuable for high-resolution imaging. However, such image streams are degraded by both motion blur and noise. Building on deep learning based approaches developed for deblurring videos of natural scenes we explore the opportunities and limitations of deblurring and denoising images captured from a fast image stream collected by a TEM microscope. We start from existing neural network architectures and make adjustments of convolution blocks and loss functions to better fit TEM data. We present deblurring results on two real datasets of images of kidney tissue and a calibration grid. Both datasets consist of low quality images from a fast image stream captured by moving the sample under the microscope, and the corresponding high quality images of the same region, captured after stopping the movement at each position to let all motion settle. We also explore the generalizability and overfitting on real and synthetically generated data. The quality of the restored images, evaluated both quantitatively and visually, show that using deep learning for image restoration of TEM live image streams has great potential but also comes with some limitations.

The abstract can be rewritten to be more meaningful. The authors should add more details about their final results in the abstract. Abstract should clarify what is exactly proposed (the technical contribution) and how the proposed approach is validated.
-Rewrote a major part of the abstract to better highlight proposed method and approach.
The paper does not explain clearly its advantages with respect to the literature: it is not clear what is the novelty and contributions of the proposed work: does it propose a new method? Or does the novelty only consist in the application?
-Made the contribution clearer in the end of the Introduction. The main contribution lies in combining modern techniques for deblurring of TEM data, where the main focus is on the application.
The contributions of the paper are not clearly identified (Section 1, last paragraph). Authors need to be claimed their contributions and justify with sufficient experimental results.
-See previous answer. We have also added the result of one additional experiment to clarify the experimental results.
Bullet your contribution at the end of the introduction section.
-The contributions are now clearly listed in the last paragraph of the introduction.
Overall, the Methodology Section needs to extend with more details in each step and I recommend adding an algorithm (pseudocode) for the proposed method.
-We added a figure of the proposed network configuration along with a more detailed description of our contributions.
Authors need to provide justifications for all the parameters setting.
-We added more details on what parameters are re-used from previous publications and what parameters are optimized for the application presented here. Details regarding training approaches are also clarified.
More scientific reasoning should be added to the experimental results' explanations.
-We have now added an extra experiment and clarified our reasoning regarding model improvements and selection of technical approaches.
Manuscript needs to be thoroughly revised and rewritten in the format of a journal publication and must be edited by a native English speaker.
-The manuscript has been thoroughly revised and edited by a native English speaker.
Please highlight the advantages and disadvantages of your method.

I need to see a comparison between the proposed method with other previous methods.
-We now argue about the problems that arise when using more classical methods, like approximating the point spread function for motion deblurring. Also, the problems with using other deep learning techniques that for instance work with time series are discussed, hence motivating why we compare "only" two different network architectures.
In results, authors should add the convergence graphs.
-Since we have so many networks trained, it would take up a lot of space to include all convergence plots and would like to leave these ones out. All networks are trained for a specific number of epochs making sure they have converged.
Do the authors employ any cross-validation scheme? Please, provide details about it.
-We now added cross validation to the manuscript to show that network performance is not restricted to a specific split of the data.

Reviewer #2:
This manuscript proposed a fast image stream based transmission electron microscopy image restoration by consideration of motion blur and noise. This work itself is interesting. However, some other problems in the manuscript are still concerned in the following: Grammar mistake in the sentence "shows that using deep learning for image restoration of TEM image live streams has great potential but also carry limitations".
-We fixed grammar mistakes and let a native English speaker revise the text.
The contribution and innovation of this work should be stated more clearly.
-We made our contributions more clear at the end of the Introduction.
The organization of this manuscript should be added to the end of the introduction.
-We added a description of the organization of the manuscript to the end of the introduction.
More details on the method should be exposed in the text.
-We added an illustration of the model and added text about the content of the components in the model, as well as an additional supporting experiment.
The flow charts of the method should be shown in the manuscript.
-See previous answer.
More evaluation metrics are suggested in the experiments.