publish-micro-image

Ivan Calandra , 2024-09-16, 17:01:45

• Introduction
• Scale of observation
• Magnification
  • Optical magnification
  • Digital magnification
  • Banning magnification?
  • Scale bar
• Resolution
  • Optical lateral resolution
  • Digital lateral resolution
  • Relationship between optical and digital lateral resolutions
  • Digital zoom
  • Optical vertical resolution and depth of field
  • Digital vertical resolution
• Scale, magnification, resolution - summary
• Processing
  • Definitions
  • Recommendations
• Saving
  • Raw vs. derived data
  • File formats and software
  • JPEG, PNG, TIFF, RAW
• Reporting and sharing
  • Minimum reporting requirements
  • Sharing
  • Instrument-specific guidelines
• How to contribute
• License
• References

Recommendations on how to publish microscope images

The releases are available and citable on Zenodo

Introduction

This good-practice document is directed at archaeologists and paleontologists working with microscopes, but anyone working with microscope images might learn a few things here.

Microscopes now deliver digital images. “Digital images are data” (Cromey 2013). However, digital images are different from analog images and from what the observer/analyst sees through the oculars. Therefore, digital images have specific properties that must be reported in publications, so that other researchers have the necessary details to assess the published images and also so that they can repeat/reproduce these images.

This document is by no means exhaustive and is limited in scope on purpose, but I hope it will help many archaeologists and paleontologists.

As a disclaimer, this document only presents my opinion, at least for now. You are free to disagree. Whether you agree or not, I would really appreciate if you could share your opinion and contribute to the development of this document, so that it can be accepted by a larger community.

Scale of observation

A microscope is obviously used to observe features of an object that are too small to be visible to the naked eye. To do so, a series of lenses magnifies the object so that the analyst sees more details. But which scale(s) is (are) appropriate and how to reach this (these) scale(s) with a microscope?

So, the first question the analyst should ask is: What is the size of the features of interest? It is important to consider the range of sizes: the size of the smallest features is important, but so is the size of the largest. This is because the larger this range, the less likely it is that all these sizes can be observed with a single set of settings. In other words, if features of different sizes are of interest, several acquisitions at different scales will probably be needed.
Finding the size(s) of the features is often done directly during observation: increasing the magnification until the features of interest are visible. This is a valid approach if the microscope used has the appropriate hardware (objectives, zoom, camera, etc.) to resolve the features of interest. But it can also happen that the features of interest are smaller or larger than what is permitted by the hardware at hand. Moreover, when dealing with digital images, some other issues might occur. This is why it is always advisable to start with at least a rough idea of the size(s) of the features and of the capabilities of the microscope(s) available to the analyst.

But how do we know what the microscope’s hardware can achieve in terms of scale? Most publications and reports talk about magnification. But what is it really and what does it mean? In the next section, I will argue that magnification is pretty much useless in digital microscope imaging and actually only confuses the analyst. Resolution is much more important, but it is unfortunately still cryptic to many analysts; I will therefore try to explain some concepts afterwards.

Magnification

Magnification is simply: the size of the magnified area divided by the real size of the imaged sample’s area.

Optical magnification

But let us first have a look at the optical magnification.
The optical magnification of a microscope image when observed through the oculars is straightforward to calculate:

$optical \ mag. = objective \ magnification \times ocular \ magnification \times optical \ zoom$ (1)

For example, using a 20 $\times$ objective with 10 $\times$ oculars and 2 $\times$ optical zoom, the total optical magnification is 400 $\times$.

Digital magnification

Things get more complicated in the digital world. It all comes down to one single question: what is THE size of the magnified area when stored on a digital image? Is it the size of the screen when you acquire the image? Is it the size of the screen when you view the image? Is it the size printed on paper, or projected to a screen? There is no single size for a digital image because it depends on the viewing medium: increase the size of the viewing medium and the magnification will increase as well, even though the image is the same. Does that mean that increasing the size of the viewing medium will improve the resolution? Unfortunately, no, as we will see below.

Let us see how the digital magnification is calculated. The problem with a digital image (as opposed to an analog image) is that many components come into play to magnify the sample: all the optical (objective, optical zoom and camera adaptor) and digital (camera sensor, viewing medium and digital zoom) components must be considered. Note that the oculars are not relevant anymore because the light goes to the camera without passing through the oculars.
This is how to calculate the digital magnification:

$digital \ mag. = \frac{objective \ magnification \times optical \ zoom \times camera \ adaptor \times viewing \ medium \ diagonal \times digital \ zoom}{camera \ sensor \ diagonal}$ (2)

For example, using a 20 $\times$ objective, 2 $\times$ optical zoom, 1 $\times$ camera adaptor, 381 mm or 15” screen diagonal (to be precise, what matters is the diagonal of the part of the screen where the image is displayed), 1 $\times$ digital zoom, and 11 mm camera sensor diagonal, the total on-screen magnification is:

$\frac{20 \times 2 \times 1 \times 381 \times 1}{11} \approx 1385 \times$

As you can see, with the same optical components (except oculars) as in the example calculation of the optical magnification above (see formula (1)), the resulting digital magnification is about 3.5 $\times$ larger than the optical magnification. So digital and optical magnifications are different!

An example I like to show is the figure 8a of Pedergnana (2020) (the paper can be read here). According to the legend, the original magnification of the optical light microscopy image on the left is 10 $\times$ and that of the SEM image on the right is 200 $\times$ (actually 260 $\times$ according to the data zone on the image itself). So different magnifications but same field of view and roughly same size on the PDF…?! Here I do not mean that Antonella Pedergnana was wrong; on the contrary, she actually did correctly report the microscopes’ settings as it should be done. She even specified in the legend that the reported magnifications are ‘original magnifications’ to highlight the fact that the images in the PDF have other (digital) magnifications. The problem is that this acquisition setting can be misleading for the non-expert reader. This example demonstrates that the concept of digital magnification is troublesome, to say the least!

Also, the formula (2) shows that calculating the total digital magnification is not so easy in practice: the camera sensor diagonal and the viewing medium diagonal are not always known.

Banning magnification?

So what now? Should we all forget about magnification in digital microscopy? Well, pretty much, in my opinion.

If the analyst has observed and analyzed while looking through the ocular, the optical magnification is very relevant and should be reported (see section Reporting and sharing). However, the image published is a digital one. This implies that the analyst and the readers/audience will observe the sample at different magnifications (and resolutions as well, see section Resolution). In some cases, the readers/audience might not be able to see what the analyst is referring to. This might result in misunderstanding and potentially even disbelief in respect to the analyst. Even when the analyst analyzed the digital image directly, the viewing medium diagonal might change for each person anyway (laptop screen vs. desktop screen vs. projected screen vs. printed…).

As a side note, the field of view of a digital image is usually smaller (to avoid vignetting) than the field of view of the image as seen through the oculars, and the contrast and clarity when looking through the oculars is usually much better than any digital image. These are other sources of differences when the analyst observed through the ocular but reports/publishes a digital image.

There is another reason for avoiding the digital magnification. When I was tasked to compare different digital microscopes (i.e. microscopes without oculars and where the image is directly and only viewed on a screen), I noticed that there were very large variations in the magnification ranges advertised by the different manufacturers. Looking at the small print, I realized that, for example, one manufacturer reported the magnification as seen on a 15” screen diagonal, while another one as seen on a 17.5” screen diagonal. This made sense because this was the size of the image as seen in the respective acquisition software packages. But this is very confusing to the buyer/user.
Just for the comparison, let us simply change the screen from 15” to 17.5” (444.5 mm) in the example above:

$\frac{20 \times 2 \times 1 \times 444.5 \times 1}{11} \approx 1616 \times$

So this makes a significant difference!

There is another similar example, namely in SEM imaging. When using an SEM, only the digital magnification is availble, but it is given relative to a reference. For example, in the case of our Zeiss EVO 25, the magnification is given relative to the Polaroid 545 format (other formats can be selected), i.e. as if printed on 4 $\times$ 5” polaroid paper. This used to be the printing format for analog images in the days before digital SEMs. Note that other manufacturers might use different references so the magnifications might not be comparable, once again, and I am not sure all users know about this reference.

Scale bar

So, if the digital magnification is useless, how can we know the size of a digital microscope image? The good old scale bar! The field of view, i.e the size of the imaged area of the sample, or rulers or anything similar will do the trick too.

One would expect that everybody knows that a scale bar (or some form of scaling) must be provided with each image. Yet, “[a]pproximately half of papers in physiology (49%) and cell biology (55%) and 28% of plant science papers provided scale bars with dimensions (in the figure or legend) for all images in the paper […]. Approximately one-third of papers in each field contained incomplete scale information, such as reporting magnification or presenting scale information for a subset of images. Twenty-four percent of physiology papers, 10% of cell biology papers, and 29% of plant sciences papers contained no scale information on any image” (Jambor et al. 2021, 3).
In sum, only about half of the papers in these fields publish microscope images with proper scaling information; this is not much! I do not expect that archaeology is any different, although I am not aware of such a survey for archaeology.

Resolution

In general, resolution refers to the level of details that can be observed. In other words, it defines the size of the smallest observable features.

There is not one single resolution but many different, complementary definitions of resolution. First, it is important, once again, to differentiate between optical and digital resolutions. And second, there are lateral (XY) and vertical (Z) resolutions.

Most terms and definitions here refer to the Fair Data Sheet Initiative, v1.2α (the website is only in German, but the Fair Data Sheet itself has an English version). I will abbreviate it “FDS”.

Optical lateral resolution

Due to aberrations from the optics and diffraction of the light, an optical system will be limited in its ability to distinguish points; the image will become blurry at the smallest scales. These effects are the basis for the calculation of the optical lateral resolution as defined by the Rayleigh criterion.

The calculated optical lateral resolution is the “minimum theoretical distance between two adjacent, barely distinguishable features of an object” (FDS §2.2.8). In other words, if two points are closer from each other than the optical lateral resolution, then it is not possible to distinguish them (i.e. they will look like one point only). They will appear as two separate points only if they are further from each other than the optical lateral resolution.

The calculated optical lateral resolution $\delta_L$ depends on the numerical aperture of the optics (= objective + optical zoom, $NA$), on the wavelength of the light $\lambda$ and on a factor $K$:

$\delta_L = \frac{K.\lambda}{NA}$ (3)

The value of $K$ is 0.61 for bright field microscopy (= “classical” light microscopes). For laser-scanning confocal microscopy, $K$ depends on the diameter of the pinhole and varies between 0.37 and 0.61 (Artigas 2011).

From formula (3) and the definition of the optical lateral resolution, it becomes apparent that the smaller $\delta_L$, the better the resolving power of the microscope (i.e. closer points will be distinguishable). This is what we usually refer to as “higher resolution”, even though the value for $\delta_L$ is smaller. Confusing, right? This is why I prefer to talk about “better/worse” resolution rather than “smaller/larger”.

The NOTE 3 of the FDS §2.2.8 states that “the optical [lateral] resolution can be achieved only under ideal conditions (including complete illumination of the pupil) on planar objects. This value is usually not reached on textured surfaces”. In archaeology and paleontology, we usually deal with textured surfaces!

Digital lateral resolution

The digital lateral resolution, or measuring point spacing, is the “sampling interval of measuring points in the measuring volume, both in X and in Y direction” (FDS §2.2.7). This is also referred to as the pixel size (although pixel size or pixel pitch can also refer to the size of the photodiodes on the detector/camera, which is a fixed value independent of the field of view) or spatial sampling.

The measuring point spacing is calculated by dividing the measuring area (FDS §2.2.1, also called field of view) by the maximum number of measuring points in a single measurement (FDS §2.1.2, also called frame size or number of pixels):

$measuring \ point \ spacing = \frac{measuring \ area}{number \ of \ pixels}$ (4)

Relationship between optical and digital lateral resolutions

So far, so good, and not too complicated. The problem is that we now have two different values for the lateral resolution. Which one should we care about? Both, of course! But what should the relationship between optical and digital lateral resolutions be?

The Nyquist criterion (based on the Nyquist–Shannon sampling theorem) states that the value for the digital lateral resolution (measuring point spacing) should be 2-3 times smaller (approx. 2.8 times smaller according to Pawley 2006) than the value for optical lateral resolution ($\delta_L$). This is necessary to digitally image with sufficient details the transition between features that are optically visible.
For example, if $\delta_L = 0.6 \ \mu m$, then the measuring point spacing should be 0.2-0.3 µm.

If the digital lateral resolution is more than 3 $\times$ smaller than the optical lateral resolution, this will result in oversampling: adjacent pixels will have the same values, so the digital image will not contain more meaningful information but simply more information. This results in larger files than necessary and generally does not improve the quality of the image. On the contrary, it can even lead to worse images because each pixel will record less signal (i.e. lower signal-to-noise ratio or SNR, Pawley 2006).
If the digital lateral resolution is less than 2 $\times$ smaller than the optical lateral resolution, this will result in undersampling: the transition between features that are optically visible will not be visible on the digital image.

There are many considerations to take into account in the analog to digital conversion process, but as Pawley (2006, 70) wrote: “If you don’t want to worry about any of this, stick to Nyquist!”

Note that in the section Optical lateral resolution, I argued that we should avoid the terms “larger” and “smaller” when talking about resolutions. But in this section, it is really about the relationship between the values of $\delta_L$ and measuring point spacing. In any case, the relationship could also be stated that way: “the digital lateral resolution should be 2-3 $\times$ better than the optical lateral resolution”.

Digital zoom

Formula (2) includes the digital zoom in the calculation of the digital magnification, but it is not part of formula (4) for the digital lateral resolution. So what happens when zooming in digitally?

Let us start with an original image 100 $\times$ 100 µm with 1000 $\times$ 1000 pixels. This results in a measuring point spacing of 0.1 µm.
The digital zoom acts like a cropping + enlarging tool: the original image will be first cropped and then enlarged to the same viewing size as the original image.
A digital zoom factor 2 $\times$ means that the original image is cropped by half in X and in Y directions, i.e. the zoomed image is 4 $\times$ smaller than the original image. This is true both for the size in µm and for the number of pixels. The zoomed image will therefore be 50 $\times$ 50 µm but it will occupy the same space on your screen as the original image (i.e. digital magnification 2 $\times$ larger). And it will have 500 $\times$ 500 pixels.
So what about the digital lateral resolution of the zoomed image? 0.1 µm, just like the original image.

All this to say that the digital zoom obviously changes the digital magnification but does not influence the digital lateral resolution.

Still, it seems that zooming in digitally helps to see more details. This is just because the details are displayed larger on the screen, not because they are better resolved. For a better digital resolution, a smaller field-of-view for the same number of pixels, or more pixels for the same field-of-view, is needed. This can be achieved by using a higher magnification objective and/or an optical zoom, or a camera with more pixels.

Optical vertical resolution and depth of field

The optical vertical, or axial, resolution is equal to the depth of field for a bright field (BF) microscope:

$\delta_A(BF) = depth \ of \ field = \frac{n.\lambda}{NA^2}$ (5)

with $n$ the refractive index of the medium; $n_{air} \approx 1$.

For a confocal microscope, the axial resolution is calculated by the optical slice thickness, which also depends on the pinhole diameter (Artigas 2011):

$\delta_A(conf) = \frac{0.64 \lambda}{n- \sqrt{n^2-NA^2}}$ (6) with a pinhole diameter $PH < 0.25 AU$ (Airy Unit)

$\delta_A(conf) = \sqrt{(\frac{0.88 \lambda}{n- \sqrt{n^2-NA^2}})^2 + (\frac{n.PH.\sqrt{2}}{NA})^2}$ (7) with a pinhole diameter $PH$ $\ge 0.25 AU$

Here again, the smaller $\delta_A$, the better the resolution is.

Digital vertical resolution

The digital vertical resolution is meaningless in case of a 2D image (even for extended depth of focus - EDF - images) but it is relevant for 2.5 and 3D images.

The digital vertical resolution of a 2.5D image acquired from a confocal microscope is better than the step size (i.e. the distance between the individual images, or slices, of a stack) because of the way it is processed (see Artigas 2011).
In short, the focal distance is constant for a given objective, so when the microscope’s objective moves up and down, different points of the sample are in focus. Each point (pixel) on each slice records the intensity of the signal (called axial response) at that location (X+Y from stage coordinates and Z from the slice’s height). The intensity increases when the focal plane gets closer to the point on the sample and decreases again when the focal plane moves further away from the point on the sample. The curve of the axial response (intensity) vs. height of the slice therefore has a maximum that can be mathematically computed (see fig. 11.6 in Artigas 2011) and that do not necessarily fall on the height of a given slice. This maximum is the value used for the height of the given pixel. Repeat the process for all pixels, and you get a 2.5D height map with a digital vertical resolution better than what the mecanics (drives or piezo) can achieve.

Unfortunately, I do not know how to calculate the digital vertical resolution of a z-stack.
If anyone knows more about it, please contribute!

Scale, magnification, resolution - summary

Magnification and resolution are different concepts and are not interchangeable. Resolution is given in a unit of length (e.g. µm), while magnification is without unit (“$\times$”).

Magnification can be understood as an aid to see the resolution: resolution will define what can be seen or not, while increasing the magnification will make the visualization of the observable details easier. This is why, in my opinion, talking about resolution is much more useful and relevant than magnification, especially in the digital world.

Scale is a fundamental concept for any observation. Magnification is often, unfortunately, used as a proxy for scale. However, magnification is quite decoupled from scale. Scale can be properly described by field of view (FOV, i.e. the size of the region observed) and resolution (i.e. the level of details visible on the optical and digital images). Magnification informs only on the FOV, but even for that, it is not very useful when most microscopes now allow for stitching in order to increase the FOV at constant magnification and resolution.

As a side note for use-wear analysts, the two approaches to use-wear analysis are called high-power and low-power, not high-magnification and low-magnification!

In summary, forget about magnification and learn about resolution!

Processing

Few images are published without some form of processing. Sometimes, processing is even necessary to create (e.g. EDF, stitching, topography reconstruction) or analyze (e.g. contrast/brightness, wavelength filters) the image. Even the raw data output by every instrument are processed to some degree.
While processing deserves a whole good-practice document on its own, I would just like to give some general advices here that will hopefully be applicable to most processing routines. But feel free to contribute for more!

Definitions

Following the definitions of Aaron and Chew (2021, 1), processing “serves to digitally transform an acquired dataset by enhancing or isolating signals of interest and/or suppressing other signals and noise that will otherwise prevent accurate analysis”. Analysis refers to “a vast set of diverse approaches to extract […] meaningful, quantitative measurements from a dataset”. Lastly, pre-processing includes the “steps that are required to be taken before image data can be properly processed […]. Such steps generally do not serve to enhance particular features in an image per se, but rather correct for imperfections commonly encountered in imaging systems.”

Here are some examples in each category when considering surface texture analysis (dental microwear texture analysis or quantitative artifact microwear analysis):

• pre-processing: topography reconstruction from a Z-stack, stitching…
• processing: leveling, form removal, thresholding, wavelength filters…
• analysis: texture analysis following ISO 25178, scale-sensitive fractal analysis, furrow analysis…

In the remaining of this section, I will use the term processing to include both pre-processing and true processing. This makes the sentences easier to read and the discussion applies to both anyway.

Recommendations

Even if we sometimes tend to assume that there is only one way to process an image, there are in fact often several methods or algorithms to do so, and there are always various settings to adjust. Additionally, some processing happens before you get the raw data from the instrument. Lastly, “[a]s images are numerical data, image processing invariably changes these data and thus needs to be transparently documented” (Schmied et al. 2023, 7). This is why it is important to be as precise as possible when reporting about the acquisition and processing (see section Reporting and sharing).

It is also crucial to process all images of a dataset in the same way. This might seem obvious, but different processing methods might produce results that might appear identical/similar even if they actually are different. Processing all images in the same way ensures that all images are comparable at least within a dataset. Of course, this is true only if the images were acquired in the same way!
So, while the details are more nuanced, I believe the message is clear and every analyst should try his/her/their best to acquire and process images and data in a consistent and comparable manner.

Another general piece of advice is to keep the raw data, in this case the acquired image, and to process a copy of it rather than overwriting the acquired image (see e.g. Cromey 2013; Schmied and Jambor 2021). This way, it is possible to check the result and process differently if necessary. See section Saving for a discussion on raw/derived data and file formats.

Last but not least, it is important to keep a very detailed record of the processing and analysis steps that were performed. Templates and scripts will make it easier to repeat and reproduce the processing/analysis. This is not only important to share data (see section Sharing); it is also important for yourself as it is likely that you will have to repeat the analysis at some point, at least for some images. protocols.io can be a useful tool for this.

Fiji/ImageJ is a great, open-source tool for image analysis. Schmied and Jambor (2021) proposed guidelines for image analysis and a processing workflow in Fiji; there is no reason not to follow and use them! The checklists of Schmied et al. (2023) are also very useful.

Saving

Raw vs. derived data

I totally agree with Ben Marwick and colleagues that raw data should be kept raw and clearly separated from derived data (e.g. Marwick and Pilaar Birch 2018; Marwick, Boettiger, and Mullen 2018).

Still, it is not always clear to distinguish between the two, nor practicable to keep the raw(est) data. For example, the raw data of an EDF image or a 2.5D topographical model (= height map, e.g. from a confocal microscope) is a Z-stack. Z-stacks can be very large files so keeping all these Z-stacks requires a lot of storage space. While storage is not really limiting anymore, the environmental impacts of data centers and online repositories cannot be ignored (e.g. Samuel and Lucivero 2020).

Also the derived data of an analysis can be the raw data of another. Using the example of surface texture analysis (dental microwear texture analysis or quantitative artifact microwear analysis), the raw data of the analysis are the height maps (which are themselves derived from pre-processing the Z-stacks) that are used as input for MountainsMap (or equivalent) to calculate the surface texture parameters for each height map, which are eventually exported into a CSV file. This CSV file is in-turn the raw data for the subsequent statistical analysis.

All this to say that the decision should be made on a case-by-case basis about what data is raw and what is derived. Nevertheless, there are some general principles that apply in most (all?) cases. Make sure to:

• save the rawest data possible, usually the acquired, unprocessed image;
• save every derived data that is included in the final analysis (i.e. input data of the results), especially if the processing/analyses are not fully automatized;
• save all the metadata associated with both the raw and derived data (see section File formats and software);
• save in appropriate format(s) (see section File formats and software).

File formats and software

For long-term usability, accessibility and sustainability, proprietary formats (i.e. files that can only be opened with a paid software) should be avoided and stable, open, non-proprietary formats should be preferred. The DANS maintains a list of preferred and non-preferred formats. For microscopy images, the OME-TIFF format has become the standard.

Most microscope manufacturers offer software packages to acquire images; in some cases, these software packages are even necessary to operate the microscopes. Not all of these software packages can save in an open format, even less so in OME-TIFF. Unfortunately, metadata sometimes get lost when the image is not saved in the native, proprietary format.
This is why, if your acquisition software cannot natively save in an open format, I would recommend to save the image in the proprietary format (in order to save all the metadata) and to additionally export to an open format (for long-term accessibility). Sometimes, it is also possible to export the metadata into an open format (e.g. JSON, XML, CSV or TXT).

Following on this discussion, it is therefore important to favor open-source software packages over paid ones. Nevertheless, open-source projects often have difficulties to keep running and being developed (see e.g. Coelho and Valente 2017). Also, open-source software packages are often less user-friendly, and are slower to add new functionalities and to adapt to OS upgrades. So, open-source is not always better.
Still, Fiji/ImageJ is a great open-source solution for image analysis with a 25-year-long history (meaning it is likely to survive). Bio-Formats is an important “software tool for reading and writing image data using standardized, open formats”, which is incorporated into ImageJ. Check the how-to ImageJ2-Fiji for details on how to read Zeiss files and access their metadata with this software.
Some microscope manufacturers also offer a free viewer or a free light version (e.g. Zeiss’ ZEN starter and ZEN lite).

JPEG, PNG, TIFF, RAW

Following up on the previous section, every one of us has wondered at least once which image format should be preferred and what the differences are between JPG, JPEG, PNG, TIF and TIFF. Here, I will try to give an overview and general recommendations. This section concerns only raster image formats, as microscope images are always raster images.

I recommend reading the documentation about the different raster file formats on the Adobe website, which is, in my opinion, clear and detailed.

• TIFF (.tif or .tiff, same format but different extensions for historical reasons) is the format to choose if you want to save images at full resolution and with best quality, which in turn implies that TIFF files are large. Metadata can be saved together with the image when using this format (e.g. for SEM images).
• JPEG (.jpg or .jpeg, same format but different extensions for historical reasons) is a widely used format for photos. Files are small and of generally good quality, but the lossy compression means that some details are lost. JPEG files can include metadata, especially the camera’s EXIF data, but is not recommended for scientific imaging (Cromey 2013).
• PNG is best knowned for their transparent backgrounds, but it is also a good format for any type of image thanks to its lossless compression. Files are smaller than TIFF but larger than JPEG. While many users tend to prefer JPEG, I think PNG could be better for many applications and should be preferred to JPEG (Cromey 2013; Schmied and Jambor 2021; Schmied et al. 2023). Compare the two formats here.
• I am not aware of any microscope camera that can save in RAW, but DSLR cameras can usually do it. So this format is worth considering for users who have a DSLR camera mounted on the phototube.

Reporting and sharing

“The underlying premise of image publication and ethics guidelines is that a digital image is data and that the data should not be manipulated inappropriately” (Cromey 2013, 5).

All the background information of the previous sections aims at providing the required knowledge to use a microscope to reach the desired scale of observation and to save all the acquired data and metadata. While it is important for the analyst to see what he/she/they want to see on the sample, it is equally important that the reader/audience that will be presented - or even will assess - the work is able to know what can be observed in terms of scale on the reported/published/shared images. This is also important for repeatability and reproducibility (and pre-producibility sensu Stark 2018).

The importance of reporting for microscopy is not unique to archaeology/paleontology (see e.g. Calandra et al. 2019) and other fields also suffer from a lack of high-quality reporting (see e.g. Marqués, Pengo, and Sanders 2020 in biomedical research).

A large community of researchers in life sciences are working on guidelines for reporting microscope data (e.g. Aaron and Chew 2021; Hammer et al. 2021; Heddleston et al. 2021; Jambor et al. 2021; Ryan et al. 2021; Sarkans et al. 2021; Schmied et al. 2023, as well as the Open Microscopy Environment - OME, the European Light microscopy initiative - ELMI, the Society for Microscopy and Image Analysis - GerBI-GMB and the Quality Assessment and Reproducibility for Instruments & Images in Light Microscopy - QUAREP-LiMi communities). The work they do is incredible, but unfortunately, it concerns mostly life sciences and some aspects are not relevant to archaeology, while some others relevant to archaeology are not addressed.

Minimum reporting requirements

Below is a list of what I think is necessary for pre-producible microscope images. This list should be adapted according to the equipment/software used.

From my point of view, all of these pieces of information should be published together with every microscope acquisition.

I would like to stress that most of these metadata are generated and saved automatically together with the image data (at least when you use the appropriate format) so that there is no extra work needed to compile the image metadata beyond sharing (see section Instrument-specific guidelines). I therefore do not see a reason why we should not report them.

Additionally, I recommend following the guidelines of Jambor et al. (2021) to improve the quality of the images themselves, as well as using the checklists of Schmied et al. (2023) to make sure that at least the relevant “minimal” requirements are met (but aim for the “ideal” requirements!).

Sharing

A lot could be said about data sharing and Open Science. I will keep it brief here and focused on microscope images; more general and extensive discussions can be found in e.g. Marwick and Pilaar Birch (2018), Gomes et al. (2022), and Karoune and Plomp (2022).

In my opinion, publishing only a few, supposedly representative images of the whole dataset is not enough because of the following reasons:

1. The sample might be representative in some aspects but not in others.
2. To be honest, in most cases, only the “nicest” images are published and the least informative/relevant are not; this implies a strong bias and the true signal might just be lost by this selective sampling/reporting.
3. Images published in a journal article or in a book are usually small and of reduced quality (in the PDF at least), so the published images might not even show what can be seen on the full-resolution image.
4. Without seeing all images in full resolution, how can anyone assess whether the data support the conclusions?
5. It would be a shame to keep all these images for yourself.
6. Why should we show only a small part of the extensive work we did? Let us do ourselves a favor and show how hard we work!

Or as Cromey (2013, 3) puts it: “If we only show other people the pictures we want them to see, then the viewers will most likely reach the interpretation that we want them to have. What if our interpretation is wrong? It may be a poorly kept secret that most published images are somewhat less than “representative”, but is this right? What would happen to the quality of science if we only showed our most compelling images to our project leaders, lab group, or collaborators?”

I would therefore recommend:

1. to keep showing a sample as representative as possible in the journal article, but
2. to additionally upload all images with their metadata to an online repository.

I would like to stress that the data should be uploaded to a trustworthy online repository (e.g. Zenodo, OSF or Figshare), rather than as online supplementary information (SI) to the article. This is because SI might be behind a paywall, just like the article. Even when it is not, publishers might have restrictions about file formats and names. And since everyone does mistakes, it is easier to correct them if you have access to the repository rather than relying on the publisher to do it (which can take a long time or might even never happen).
Ideally, data should be uploaded to a discipline-specific repository rather than a generalist repository like Zenodo or OSF. But I am not aware of any archaeology-specific repository that is commonly used. I have heard about for example The Digital Archaeological Record (tDAR) but it seems to be mostly used for excavation data. Please contribute!

Here again, the discussion about raw/derived data and file formats is important (see section Saving): both raw and derived data should be shared in formats that include all the metadata (Schmied et al. 2023).

I propose step-by-step instructions to upload files to Zenodo in how-to Zenodo.

Instrument-specific guidelines

I have tried to make it easy for the users of the Imaging Platform at LEIZA (IMPALA): a Shiny App guides the users to fill in the required information and exports the report to an XLSX or ODS file that can be shared together with the images (see section Sharing). This App is still in development and is part of the repository imaging-reports on GitHub.

To complement this, I have prepared guidelines specific to the instruments we have at the IMPALA in the folder Guidelines. The guidelines include text snippets that can be copy/pasted/edited for the method section of a publication and have been converted from Markdown to DOCX format using the extension vscode-pandoc. They also explain in details what type of data should be shared and how.
Feel free to contribute guidelines for other instruments!

Neither the reports nor the text snippets are meant to be used for every acquisition. Instead, they serve to summarize the acquisition and analysis processes and should be used only once for each publication/project.
But this works well only in combination with the detailed metadata shared together with the data themselves, as detailed in the guidelines (see also section Sharing).

How to contribute

I appreciate any comment from anyone (expert or novice) to improve this document, so do not be shy! There are three possibilities to contribute.

1. Submit an issue: If you notice any problem or have a question, submit an issue. You can do so here.
   A GitHub account is necessary.

2. Propose changes: This document is written in Quarto. If you know how to write in this format, please propose text edits as a pull request (abbreviated “PR”).
   A GitHub account is necessary.

3. Send me an email: If you do not have a GitHub account and do not want to sign up, you can still write me an email (Google my name together with my affiliation to find my email address).

By participating in this project, you agree to abide by our code of conduct.

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

See also License file in the repository.

Author: Ivan Calandra

License file, badge and image from Soler (2022).

References

<div id="refs" class="references csl-bib-body hanging-indent" entry-spacing="0">