From pixels to length: Body length estimation of aquatic macroinvertebrates from digital images for ecological applications

Abstract

Accurate measurement of macroinvertebrate body length is a key component of ecological research, providing essential data for estimating biomass, evaluating growth rates, and understanding trophic interactions in aquatic ecosystems. Yet, comprehensive datasets remain scarce, largely due to the labor-intensive and error-prone nature of traditional manual measurements – especially when processing bulk, mixed-species samples, where body sizes of individual specimens may range over orders of magnitude. This study presents a streamlined, non-destructive approach using Fiji, a free open-source image processing software, in combination with the Labkit plugin to extract body length information from digital images. Its novelty lies in enabling high-density bulk sample processing, interactive segmentation to overcome computational limitations, and taxon-specific area-length regression within a reproducible and transparent end-to-end protocol. In this workflow, body length is manually measured for a representative subset of specimens, while body area is extracted from segmented images for all. A linear regression of body length (Y) on image-derived area (X) allows reliable length estimation for specimens not measured manually. This approach improves throughput, offering an accessible and scalable solution for large-scale ecological assessment.

• Semi-automated workflow using Fiji and Labkit.

• Regression-based estimation of body length from image-derived area.

• Fast, reproducible alternative to traditional manual measurements.

Graphical abstract

Specifications table

Subject area	Environmental Science
More specific subject area	Freshwater biomonitoring, morphological trait analysis, image-based macroinvertebrate measurement
Name of your method	Image-based segmentation and morphometric measurement of macroinvertebrates using Fiji and Labkit
Name and reference of original method	[1] C.A. Schneider, W.S. Rasband, K.W. Eliceiri, NIH Image to ImageJ: 25 years of image analysis, Nat. Methods 9 (2012) 671–675. https://doi.org/10.1038/nmeth.2089 [2] M. Arzt, J. Deschamps, C. Schmied, T. Pietzsch, D. Schmidt, P. Tomancak, F. Jug, LABKIT: Labeling and Segmentation Toolkit for Big Image Data, Front. Comput. Sci. 4 (2022) 777,728. https://doi.org/10.3389/fcomp.2022.777728
Resource availability	Fiji (ImageJ distribution): Fiji Downloads Labkit plugin: Labkit - Intuitive Pixel Classification in Fiji

Background

Large-scale ecological monitoring projects often generate extensive collections of macroinvertebrate specimens across numerous sampling sites and events. These bulk samples are usually taxonomically diverse with various species and developmental stages covering a wide range of body sizes, which is technically understood as a linear body dimension, e.g., body length or width of a sclerotized head capsule. Accurate measurement of body size, particularly body length, is essential for estimating key ecological parameters such as biomass and secondary production. Traditionally, body length measurements were performed manually on preserved specimens using rulers, calipers, dissecting microscopes, or ocular micrometers. While these approaches are practical and well suited for small-scale studies or foundational work, they have clear limitations when applied to large-scale monitoring involving large numbers of individuals. Manual measurements are time-consuming and labor-intensive, particularly when processing large numbers of specimens [[3], [4], [5]]. In addition, extensive handling can damage fragile specimens, limiting their further use for additional analyses or long-term archiving. Furthermore, they are not easily repeatable or verifiable due to the lack of permanent digital records. To overcome the limitations and meet the increasing demand for large-scale and reproducible ecological data, more recent developments have shifted toward non-invasive, image-based, and semi-automated methods, which can help improve efficiency and consistency [3]. While different image-based and machine learning approaches exist, fully standardized workflows for bulk macroinvertebrate samples remain scarce, and many require large annotated datasets or high computational resources, limiting accessibility in routine ecological monitoring. In addition, many deep learning approaches require iterative model tuning, validation, and technical expertise, which may limit both their accessibility and reproducibility across studies and users.

In our study, we adapted a previously developed image analysis protocol based on Fiji, a distribution of ImageJ [6] to fit our specific needs. The plugin, the original workflow relied on, struggled to handle the high density of individuals, taxonomic diversity, and variable image quality typical of our bulk macroinvertebrate samples. We often faced computer crashes, memory overload issues, or inaccurate object detection, mostly due to large sample sizes - with some images containing up to 800 individual specimens in a single frame. As a solution, we integrated the Labkit plugin, as it performs segmentation (the process of distinguishing objects of interest from the background) in real-time and in an interactive way, so segmentation errors can be corrected while training without having to restart the whole process. This drastically reduced computational load, making it suitable to process complex images without memory issues on a standard laptop (8 GB RAM is typically sufficient). The ability to make manual adjustments immediately improved segmentation accuracy and significantly reduced the time spent on this workflow step. The key output of segmentation was a binary image with clearly separated and labeled individuals from which surface area could be extracted automatically. After estimating individual surface area, we manually measured body lengths for a subset of individuals to establish taxon-specific linear regressions between the measured lengths and the extracted area. This enabled us to obtain body length data automatically for many more specimens in the same or other samples.

This semi-automated workflow provides a robust basis for further ecological studies requiring large-scale body length data, as it significantly reduces processing time and effort while preserving specimen integrity. Rather than introducing a new algorithm, this study presents a fully reproducible, step-by-step workflow integrating Fiji and the Labkit plugin, specifically adapted for high density, taxonomically diverse bulk samples. The combination of interactive segmentation and taxon-specific regression-based length estimation provides a practical and scalable solution that, to our knowledge, has not been described in detail before. Advanced segmentation approaches (e.g., deep learning) exist, but they often require substantial annotated datasets and computational expertise, limiting their accessibility for routine monitoring workflows. In practice, this workflow enables body length estimation for the majority of individuals (∼80–95%) based on manual measurements of only a small subset (∼5–20% per taxon), substantially reducing overall measurement effort in large datasets. Although formal benchmarking was not performed, the workflow substantially reduces manual measurement effort and enables efficient processing of high-density samples in practice.

Method details

Digital image analysis is widely used in ecological studies, for example, to estimate plant cover [7] or to assess marine biomass [8]. However, standardized and detailed protocols for measuring the body length of macroinvertebrates - particularly from bulk samples using digital images - remain limited. This protocol aims to fill that gap by providing a clear and accessible workflow (for steps see Fig. 1) for researchers, including those without prior experience in photography or image analysis.

Fig. 1.

Schematic overview of the image processing and body length estimation workflow. Blue boxes indicate the main steps, with accompanying supplementary explanations in light orange. Pink highlights the central aim of the workflow, and green box indicates the novel element introduced in this protocol.

Images were acquired using a standard digital camera under consistent lighting conditions and with a uniform, high-contrast background to facilitate segmentation. A reference scale was included in each image and used for spatial calibration in Fiji prior to analysis.

Image resolution should be sufficient to clearly resolve specimen boundaries; in practice, resolutions of at least ∼3000–5000 pixels along the longest image axis were found to be adequate for reliable segmentation of bulk samples. In general, each specimen should be represented by a sufficient number of pixels (typically, >50–100 pixels per individual) to ensure accurate boundary detection.

Importantly, the workflow does not require specialized imaging equipment or high-performance computing resources and can be implemented using standard laboratory setups and commonly available personal computers. The method is suitable for processing large sample sizes and includes detailed step-by-step instructions to ensure reproducibility. Although the workflow does not yet incorporate automated specimen recognition, it lays the groundwork for future approaches in taxonomically resolved, image-based community biomass estimation. The method includes:

• •Best practices and troubleshooting tips for image acquisition: optimal setup (e.g., background, focus) and solutions to common issues such as glare, low contrast, or overlapping specimens.
• •Image segmentation and morphometric analysis using Fiji and the Labkit plugin, including surface area extraction.
• •Training and application of an interactive pixel classifier (Labkit), including iterative refinement of foreground and background annotations to improve segmentation accuracy.
• •Manual body length measurements to build area-length regression models.
• •Estimation of body length of individuals not manually measured, based on image-derived surface area using regression equations.

Required software

Fiji

Download: https://imagej.net/software/fiji/downloads

Fiji is an open-source image processing package broadly used in biological research. It includes built-in support for many plugins, including Labkit [1,9].

Fiji installation:

• 1.Download the version that matches your operating system.
• 2.Extract the ZIP file.
• 3.No installation needed - Launch Fiji by clicking Fiji application.
• 4.To ensure compatibility with Labkit, make sure Fiji is up to date. For this: open Fiji, go to Help > Update > Apply changes > OK and then restart Fiji.

Note: To create a shortcut, right click on Fiji application, choose Send to and then Desktop (create shortcut).

In our study, all image processing was performed in Fiji (ImageJ 1.53t, Java 1.8.0_322, 64-bit) using the Labkit plugin (installed via the Fiji update site: https://sites.imagej.net/Labkit/).

Labkit (Fiji) plugin

Labkit is an intuitive, real-time segmentation tool that can efficiently process high-resolution and dense images without memory issues, even on a standard laptop [2].

Download: https://imagej.net/downloads

Labkit installation in Fiji:

• 1.Open Fiji.
• 2.Go to Help > Update.
• 3.Click Manage update sites, then scroll down and enable “Labkit”.
• 4.Click Apply and Close, then Apply Changes and close the ImageJ Updater window.
• 5.Restart Fiji and Labkit will appear under: Plugins > Labkit.

Segmentation is performed using an interactive pixel classification approach, where users annotate representative regions of specimens (foreground) and background. The classifier is trained iteratively, and segmentation results can be refined in real time. Default feature settings provided by Labkit were used, allowing a balance between usability and segmentation performance without requiring parameter optimization.

All image processing steps were performed on a standard personal computer (e.g., 8–16 GB RAM), demonstrating that the workflow can handle high-density samples without the need for high-performance computing infrastructure.



Materials:

• •Digital camera (e.g., Canon EOS 2000D)
• •Tablet (to control the camera remotely)
• •Remote control software (for the tablet-based camera operation)
• •Lamp (for constant, diffuse lighting)
• •Petri dishes (for arranging specimens)
• •Reference scale (for calibration)
• •Tripod (to stabilize the camera during imaging)
• •Computer or laptop (with Fiji and Labkit plugin installed)
• •(Recommended): Acrylic (PMMA) base plate, as it ensures even illumination and reduces shadows and optical distortion [10], thereby improving the clarity of body outlines and enabling more accurate segmentation.

The listed equipment represents a typical setup; however, equivalent alternatives can be used, as the workflow is designed to be flexible and does not depend on specific hardware configurations.

Photographing

Photographing is a crucial step in this protocol, as the accuracy of the final body length dataset, as well as the entire image processing workflow is directly influenced by the quality and consistency of the captured images. Minor factors - such as background color or specimen positioning - can have a significant impact on segmentation efficiency.

This section outlines a streamlined imaging setup rethought through extensive trial and error, incorporating lessons learnt from mistakes (e.g., poor contrast, overlapping specimens). Following the steps described here can substantially reduce the need for manual corrections during image analysis and improve reproducibility across studies or sampling events.

Step 1: Pre-processing of macroinvertebrate samples

• • Sorting (optional)

  If the study requires taxonomic resolution (e.g., genus, family, or order), specimens should be sorted into distinct taxonomic groups prior to imaging (in our study, each sample was divided into 18 taxonomic groups).

• • Transfer for imaging

  Distribute specimens evenly into a clean, transparent imaging container – preferably a glass Petri dish.

  Note: The size of the Petri dish is not critical and should be selected based on the size of the sample or organisms, as long as a reference scale is included in the image. However, using a larger Petri dish is recommended, as it facilitates easier image cropping.

• • Subsampling (if needed)

  For overcrowded taxa, where the number of specimens exceeds the Petri dish capacity, subsampling is necessary to improve image clarity. A practical approach is as follows:

• ○ Spread the group evenly across a tray or Petri dish placed over a visible grid (e.g., a white tray with marked quadrats or a sheet of graph paper placed underneath).

• ○ Randomly select a subset of quadrats for imaging.

• ○ This helps minimize spatial bias by ensuring full-surface coverage.

• ○ Aim to include a minimum of 50–100 individuals per image to ensure representative measurements. If fewer individuals are captured, additional quadrats can be added.

Step 2: Image Acquisition Procedure

To ensure consistent, high-quality images suitable for digital analysis, the following settings should be optimized:

Camera setup

• 1.Mount a digital camera vertically above the imaging area using a stable tripod or a copy stand (Fig. 2.A).
• 2.Use a remote shutter release or control software to avoid movement and enable image preview (in our study, we used Canon Camera Connect).

  Background and lighting

• 3.Select a contrasting background (white or black) to maximize specimen visibility.

• ○ White backgrounds are typically better for well-saturated taxa (e.g., Ephemeroptera, Plecoptera).

• ○ Black backgrounds may enhance visibility for transparent or pale Crustacea but can obscure darker specimens such as some Diptera larvae.

• ○ For samples containing mixed-colored specimens, the background color should be adjusted as needed to achieve optimal contrast.

• 4.Position diffuse light sources to minimize shadows and surface reflections. Avoid glare caused by wet specimens.

  Note: In our study, prior to photographing, specimens were preserved in 96% ethanol and were not dried for imaging. This increased surface reflections in the image; therefore, lighting should be adjusted to minimize glare and ensure clarity.

  Imaging preparation

• 5.Place the container with the specimens centered on the background.
• 6.Distribute individuals evenly within the dish, avoiding overlaps and touching specimens.

  Note: Overlapping or touching may result in merged outlines during segmentation.

• 7.Ensure specimens do not touch the dish edges to avoid misclassification as background.

  Tip: Later, if needed, use image analysis software (e.g., Fiji’s pencil tool) to separate connected specimens or correct misclassified regions.

  Specimen-specific adjustments

• 8.Remove external structures from case-building taxa such as Trichoptera. Cases can interfere with measurements, as they do not reflect actual body size and are highly variable depending on the materials used.

  Inclusion of metadata

• 9.Place the sample identification label (e.g., site name and sampling date) within the image frame.
• 10.Include a size reference (e.g., scale bar or measuring tape – see Fig. 3.B) in each image for calibration. - In our setup, we used white A4 paper with a preprinted 2 cm line, serving as both background and reference (Fig. 2.B).

  Camera configuration

• 11.Fix camera height and angle for all images to ensure consistency in magnification and framing.
• 12.Use manual focus and live view to ensure sharpness.
• 13.Check and fix exposure and white balance settings before image capture.
• 14.Review each image immediately for quality control.
• 15.Save images using a clear, standardized file naming system (e.g., SampleID_SamplingDate) (Fig. 2.B).

Fig. 2.

(A) Setup of the tripod, camera, and sample during image acquisition. (B) Recommended arrangement of macroinvertebrates in a Petri dish, with an included label (SampleID and SamplingDate) and scale bar for reference.

Fig. 3.

(A) Ideal image acquisition setup, including tripod, camera, and a transparent platform. (B) Ideal method for including a size reference in images, using a measuring tape placed adjacent to the sample (Images taken by E. Estévez Cano).

Note: Ideally, a transparent platform (e.g., methacrylate) is used with a white background positioned approximately 20–25 cm below the camera - reduce reflections and shadows (for ideal setup see Fig. 3.A).

Image segmentation and morphometric analysis using Fiji and the Labkit plugin

Accurate image analysis - and thereby precise segmentation - is fundamental for extracting reliable key morphological metrics from macroinvertebrate specimens. Although standardized, high-quality, high-contrast images are ideal for efficient processing, such optimal conditions are not always achievable in practice. In such contexts, high throughput becomes critical. This protocol combines manual and automated techniques using Fiji, and to further enhance efficiency, we integrated the Labkit plugin, which introduces intuitive, real-time segmentation. Labkit allows interactive correction on the fly - saving time on restarting the process after adjustments, decreasing computational time even for suboptimal or artifact-prone images, and performs well even on dense images. The workflow was refined to address common challenges such as overlapping specimens, low contrast, and reflections. By following the calibration, cropping, and segmentation steps described here, researchers can obtain reproducible, high-throughput results suitable for both individual-level measurements and large-scale ecological datasets.

Image Processing Workflow overview

The following steps summarize the core procedure for processing macroinvertebrate images in Fiji and extracting morphometric measurements and are intended to aid understanding of the full workflow.

Step 1: Image preparation: Calibration serves to ensure accurate area and length measurements, the pixel-to-length ratio of each image must be calibrated based on a known distance. This step is essential for converting image data into real-world units (e.g., mm or cm). Cropping removes unnecessary areas from the image (e.g., labels, unused corners) and focuses the analysis on the region of interest. During this process, ensure that relevant metadata (e.g., country, site code, date, taxon name) is included in the file name to maintain traceability.

Step 2: Image conversion and segmentation: After the region of interest has been cropped, the image must be thresholded, converted into a binary (black and white) format for analysis, which serves as the basis for segmentation - the process of isolating individual specimens. It is essential that the image does not contain specimens touching the edges or overlapping with each other, as this may result in inaccurate measurements or exclusion during processing. Such issues should be corrected prior to analysis. Depending on the specimen characteristics and image quality, we present two different segmentation methods in this section: Thresholding, which is suitable for images containing a small number of uniformly colored specimens that exhibit high contrast against the background. Segmentation (using Labkit plugin), which is recommended for more complex images, containing diverse - with multiple genera, varying colors, sizes - or abundant specimens, where manual thresholding is unreliable.

Step 3: Extracting measurements in Fiji: Once the segmented image has been exported from Labkit, and the necessary corrections are done, all necessary visual information is available to analyze particles and extract the morphometric measurements. This step generates key size and shape descriptors (e.g., area, perimeter, body length). Proper configuration of the measurement settings ensures consistency across specimens and facilitates reproducible and standardized data.

Step 4: Data export: The previous step generates three key outputs: a labeled image displaying numbered specimens, a Results table with the extracted measurements, and the ROI Manager containing all segmented outlines. All three need to be exported from Fiji.

Keep the original image file name for each row to enable easy matching with sample metadata during analysis.

Step 1: Image preparation - Calibration and Cropping

To calibrate the Image Scale:

• 1.Launch Fiji.
• 2.Open the image either by dragging it into the workspace (the grey area beneath the toolbar – see Fig. 4) or by navigating to File > Open, selecting the desired image file, and clicking Open.
• 3.From the toolbar, select the Straight Line tool. (Fig. 5).

  (Tip: Right-click the tool icon to access additional drawing options, if not immediately visible.)

• 4.
  Locate the size reference in the image (e.g., a printed scale bar) and draw a straight line over the known distance using the mouse.

  • •
    Hold Ctrl + Shift while drawing to ensure a perfectly straight line.
  • •
    For more precise placement of the endpoints, zoom in by holding Ctrl and scrolling with the mouse wheel.
  • •
    The line’s position can be fine-tuned by clicking and dragging the endpoints (see Fig. 6 for illustration).
• 5.Go to Analyze > Set Scale.
• 6.
  In the dialog box that appears:

  • •
    Enter the Known Distance (e.g., 2.0 if the reference bar is 2 cm long and the Unit of Length is set to cm).
  • •
    Specify the Unit of Length (e.g., cm or mm).
  • •
    Optionally, check the Global box to apply this calibration to all images opened during the session.
    This setting remains active until the scale is manually changed or the Fiji software is closed.
  • •
    Click OK to confirm and apply the scale calibration (see Fig. 7).

Fig. 4.

Fiji workspace interface - images can be opened by dragging and dropping them into this window.

Fig. 5.

Fiji toolbar with the “Straight Line” tool and its configuration options.

Fig. 6.

Recommended placement of the calibration line on the scale reference.

Fig. 7.

“Set Scale” dialog box in Fiji, where the known distance and the unit of length are defined. Optionally, check the “Global” box to apply the settings across all images.

To Crop the Image to the Region of Interest:

• 1.Select the Rectangle tool from the toolbar.
• 2.Use the mouse to draw a square around the area containing the macroinvertebrates (see Fig. 8.A) – for increased precision, zoom in using Ctrl + mouse wheel.
• 3.Once the desired region is selected, go to Image > Crop.
• 4.This action keeps only the selected portion of the image - we will work with - and discards the rest. (see Fig. 8.B).

Fig. 8.

(A) selected region of interest from the original image. (B) Cropped output used for further processing.

Step 2: Image Conversion and Segmentation

Method 1 –Thresholding method for Uniform Specimens

• 1. To simplify the cropped image, convert it to grayscale.

Navigate to Image > Type > 8-bit (see Fig. 9).

• 2. Apply thresholding to create a binary image:

Fig. 9.

(A) Original color image of the specimen. (B) Converted grayscale (8-bit) version prepared for thresholding and binary conversion.

Go to Image > Adjust > Threshold.

This function converts the grayscale image into a binary (black and white) format. Adjusting the threshold range using sliders allows you to define which pixel intensity values are interpreted as foreground (e.g., black) and background (e.g., white).

Note: Automatic thresholding may fail to include pale or shiny body parts (e.g., due to pigmentation or reflection), as shown in Fig. 10.A. In such cases, manual adjustment of the threshold range is recommended (Fig. 10.B).

Fig. 10.

(A) Result of automatic thresholding, where reflective or lightly pigmented areas on the legs are excluded. (B) Suggested regions for manual threshold adjustment to enhance detection accuracy.

Be cautious not to over-adjust, as extending the threshold range may artificially inflate the object outlines, distorting true shapes (e.g., widening legs) and reducing measurement accuracy.

Limitations:

This method is only recommended when thresholding produces accurate and clean specimen outlines (see Fig. 11) without the need for excessive correction.

Fig. 11.

Examples of well-thresholded images suitable for subsequent analysis.

Method 2 – Real-Time Segmentation Using Labkit for Diverse Specimens

Additional note on alternative segmentation – Weka Trainable Segmentation plugin:

The Weka Trainable Segmentation plugin was initially tested during preliminary phases of the study. It produced promising results and has been successfully applied in similar research contexts. However, in our specific case – due to the complexity and variation of specimen appearance, image quality, and limitations of the available computer hardware – Weka Trainable Segmentation presented persistent memory warnings and slow processing times. These constraints made it less suitable for our workflow, particularly when processing larger datasets.

For these reasons, we opted for a lighter, faster segmentation method that maintained reliability without requiring high computational capacity. This choice ensured consistent performance on standard laptops and workstations.

• 1. Open the cropped image in Labkit:

Go to Plugins > Labkit > Open Current Image with Labkit (see the result in Fig. 12).

• 2. Annotate background and foreground:

Fig. 12.

Cropped specimen image loaded into the Labkit plugin within Fiji.

Once the image is displayed, select the „Draw” tool (pencil icon). Use two distinct colors to manually label representative background (e.g., blue) and foreground/specimens (e.g., red) regions.

• •Brush size can be adjusted (ideal thickness is between 1–5) (see Fig. 13.A) but note that thicker brush may increase processing time.
• •You may draw lines on and along the specimen boundaries (see Fig. 13.B) or simply place representative dots in the distinct background and foreground areas (see Fig. 13.C).

Fig. 13.

(A) Labkit interface showing the ”Draw” tool (pencil icon) and brush size adjustment bar. (B) Zoomed-in view illustrating the use of background (blue) and foreground (red) lines drawn around and on the target regions. (C) Alternative annotation method using background (blue) and foreground (red) dots placed on representative areas after zooming in Note: Segmentation lines and dots were manually enhanced for illustration purposes to improve visibility.

Zooming in with the scroll wheel can improve accuracy during annotation.

• 3. Train the classifier:

Click Segmentation > Train Classifier. Labkit will segment the image based on the labeled areas.

After the initial segmentation, review the result to ensure that annotations are accurate, there are no segmentation errors (for example, see Fig. 14).

Fig. 14.

Examples of typical segmentation errors encountered during classifier training.

Note: A key advantage of Labkit is its real-time segmentation feedback, which enables immediate visual assessment and simultaneous corrections. This interactive process allows the user to refine the segmentation as it progresses, making the workflow much more efficient.

• 4. Toggle segmentation visibility:

Use the visibility toggle (eye icon; see Fig. 15.A) to switch between the original and segmented views.

Fig. 15.

(A) Toggling visibility to alternate between the original image and the segmentation overlay. (B) Example of a correctly segmented image after successful classification.

Adjustments can be made directly on the segmented image. You may redraw incorrectly annotated areas and retrain the classifier as needed until the segmentation output is satisfactory (see Fig. 15.B).

Draw > Background and Foreground (as detailed in point 2; Annotate background and foreground), followed by Segmentation > Train Classifier.

Note (optional step to enhance processing speed):

Before exporting the segmented result, the trained classifier can be saved and reused on images captured under similar settings. This can significantly reduce segmentation time in future analyses.

• - Once segmentation is complete and the result is satisfactory, go to Segmentation > Save Classifier (Fig. 16).

• - To reuse the saved classifier: open a new image in Labkit and go to Segmentation > Open Classifier to load the previously saved model. The saved settings will be applied instantly.

  Note: Not all classifiers are universally applicable. Differences in specimen morphology or lighting conditions between images may produce incorrect results. Additionally, if the saved classifier does not suit the characteristics of the new image, Labkit does not provide an option for manual correction of the result. In such cases, either select a more appropriate saved classifier or perform manual segmentation.

• 5. Export the segmented result to Fiji and apply manual corrections if necessary:

Fig. 16.

Saving a well-segmented image is a crucial step when the same segmentation should be applied to other similar images for consistent and faster analysis.

Export the result to Fiji via Segmentation > Show Segmentation Result in ImageJ.

This creates a binary mask of the segmented image that can be manually refined/corrected in Fiji if necessary, to get a clear, well-segmented binary image (see Fig. 17.A-D).

Fig. 17.

Image showing: A) Binary image exported from Labkit. B) Location of the „Color Picker” tool. C) Open color palette with the double-headed arrow for toggling between black and white. D) Example of a clean, well-segmented binary image.

Tools to use for manual corrections:

• - Brush or Pencil tool (Fiji toolbar): to adjust the tool size, double-click the icon. A width of 5–6 pixels is suitable for minor corrections.

• - Color Picker: to choose the suitable correction color (here black and white).

• Double-click on the Color Picker icon in the toolbar to open the color palette or type „Color Picker” in the search bar, select it, then click Run.

  To toggle between black and white – depending on whether you are removing or filling regions - use the double-headed arrow at the lower-right corner of the color palette.

The following manual corrections can be applied to refine a binary image:

• - Removing unwanted fragments from analysis, by using the Brush tool to paint over them in black.

• - Connect separated body parts, by using the Pencil tool - white color with a width of at least 2 pixels.

• - Separating overlapping or touching individuals or erasing them, by using the Pencil tool - black color with a width of at least 2 pixels. An example of manual correction of clustered/touching individuals using the Pencil tool is provided in Supplementary Fig. 1.

Perfect accuracy is not required at this stage, as further thresholding or object definition will later determine final measurement values.

Important: Exported images from Labkit lose their original filenames. Rename and save them immediately (e.g., File > Save As > TIFF/PNG/JPEG) to ensure traceability.

EXTRA notes / recommendations:

If segmented images need to be saved for documentation or further analysis, the choice of file format can impact both image quality and storage requirements:

• - TIFF is the best for scientific analysis, as it is the best for preserving details.

• - PNG is good for lossless compression and is better than JPEG for clarity.

• - JPEG should only be used if file size is a concern, as it loses quality.

  In our study, segmented images were saved in JPEG format. This choice was based on practical considerations: the images were used primarily for visual reference following the measurements and specimen re-identification.

Given the high volume of material – nearly 1.5 million aquatic macroinvertebrates and several thousand high-resolution image files – the reduced file size offered by JPEG allowed for more efficient storage and handling without compromising our workflow objectives.

Step 3: Extract measurements in Fiji

• 1. Set up measurement parameters (initial setup per session)

Navigate to Analyze > Set Measurements and check the following boxes (see Fig. 18.A):

• i.“Area”
• ii.“Shape Descriptors”
• iii.“Perimeter”
• iv.“Fit Ellipse”
• v.“Display Label”
• vi.“Add to Overlay”

Fig. 18.

(A) In the Set Measurements window, select the necessary checkboxes to obtain the desired measurement parameters. (B) In the Analyze Particles window, adjust the minimum size threshold to exclude non-specimen artifacts, and ensure the appropriate boxes are selected for optimal output.

Then click OK to save the settings.

• 2. Measure objects using Analyze Particles

To extract morphometric data automatically:

• A. Go to Analyze > Analyze Particles.

• B. Adjust the settings (see Fig. 18.B):

• i. Size: 0.3-Infinity (set the minimum size to exclude small noise; 0.3 is a recommended default)

• ii. Circularity: 0.00–1.00

• iii. Show: “Outlines”

• iv. Options: “Display results”

•  • Check Include Holes only if internal holes (e.g., due to glare or reflection) appear inside the specimen (see Fig. 19. A, B).

•  • Do not check Include Holes if gaps are external or between appendages (e.g., legs or antennae crossing body) (see Fig. 19. C, D).

Fig. 19.

(A) Enabling Include Holes in the Analyze Particles window (B) may lead to overestimation of area, as enclosed spaces – such as those between legs or caused by glare – are also measured. (C) Disabling Include Holes (D) provides more accurate area values; however, any regions on the specimen excluded due to glare or reflections will not be counted.

Note: Choosing Include Holes incorrectly may result in over-segmentation or underestimation of area. If uncertain, fill problematic holes manually with white before running particle analysis.

Finally, verify that the number of detected particles matches the expected number of specimens. If discrepancies arise, adjust the binary image manually (e.g., using the brush tool) and repeat the particle analysis.

Step 4: Exporting output

Running “Analyze Particles” produces three outputs (Fig. 20):

• •A labeled image, with number tags for each specimen (if Display label and Overlay are selected).
• •A Results table, listing each specimen with area, perimeter, circularity, and major/minor axes.
• •A ROI Manager list of all segmented outlines, which allows for review, exclusion, and manual editing.

Fig. 20.

Running Analyze Particles generates three outputs: a labeled image displaying numbered specimens, a Results table with the extracted measurements, and the ROI Manager containing all segmented outlines.

Note: If Overlay is not selected, saved images will not retain labels, making future matching difficult – especially in mixed-taxa samples.

Exporting the outputs:

• •To save the labeled image: in the Fiji window choose File > Save As > JPEG or GIF.
• •To save the results table: in the Results window, choose File > Save As > All files (the file will be saved in .csv format).

Manual measurement of body length in fiji (on a subset of individuals)

While Fiji’s Analyze Particles tool provides automated estimates of length-related parameters, note that this measurement does not always reflect the true length of macroinvertebrates. For this reason, manual body length measurement remains a critical step for many groups. This section details how to accurately perform manual length measurements in Fiji, ensuring that each taxon is measured consistently based on morphological landmarks. Special emphasis is placed on defining “body length” appropriately for major macroinvertebrate groups, which is essential for standardization and comparability across studies.

Procedure to accurate manual measurement of body length:

• 1)Ensure scale is calibrated (for details, see above Calibrate the Image Scale).

  Note: If lighting, magnification, or focus differs across images, the scale must be set individually.

• 2)
  Select the appropriate tool:

  • •
    Use the Straight Line tool for linear specimens.
  • •
    Use the Segmented Line tool for curved individuals (right-click on the line tool icon to see options).
• 3)Trace the body length line: Zoom in (Ctrl + Scroll) for precise placement of the start and end points. Draw a line from the tip of the head to the tip of the abdomen – excluding antennae, cerci, or tail filaments, or between other consistent anatomical landmarks defined for body length measurement.

  For a perfectly straight line, press Ctrl + Shift while drawing the line.

• 4)Press M or go to Analyze > Measure to record the length. A Results window will appear displaying the measured Body Length (in calibrated units).

  Only Length value from this table should be used. The other values (e.g., area, perimeter) correspond to the drawn line itself – not the specimen – and are not relevant for body length measurements.

• 5)Repeat for all specimens and save results manually or save Results file as a .csv.

Body length definitions by taxonomic group

Body lengths were measured manually on binary images in Fiji, using the straight-line or segmented-line tool and following standardized anatomical landmarks. Visual examples (Fig. 21) illustrate how measurements were taken for each taxonomic group.

Fig. 21.

Images showing the measurement methods applied for different groups: Insect, soft-bodied taxon (e.g., Diptera), Decapod, Amphipod, Gastropod and Bivalve.

For general measurements, body length was defined as the linear distance from the anterior margin of the head capsule to the posterior end of the abdomen, excluding appendages such as antennae and cerci, following the method of Burgherr & Meyer [11] Though, this body length definition cannot be applied to all taxa, so the measurement approach must be varied by taxonomic group:

• 1.Insects (Plecoptera, Ephemeroptera, Megaloptera, Odonata, Trichoptera, Coleoptera, Hemiptera/Heteroptera, Lepidoptera*) and soft-bodied taxa (Araneae, Diptera, Hirudinea, Turbellaria, Nematoda, Oligochaeta*).

  Body length was measured from the anterior margin of the head capsule to the posterior tip of the abdomen, excluding antennae and cerci, where applicable.

  Note: Lepidoptera and Oligochaeta were excluded from the present study.

• 2.
  Crustacea - due to diverse morphologies of the group, there are different measurement techniques:

  • •
    Decapods: Total body length refers to the straight-line distance from the tip of the rostrum (snout) to the end of the telson (the central part of the tail fan) [12].
  • •
    Amphipods and Isopods: Total body length was measured from the anterior margin of the head to the posterior end of the uropods or telson [13].
• 3.
  Mollusca - due to diverse morphologies of the group, there are different measurement techniques:

  • •
    Gastropods: Shell length (height) was recorded from the apex to the aperture’s outer lip [14].
  • •
    Bivalves: Length was measured from the anterior to the posterior shell margin along the longest axis [15].

Optional step: body length proxies using automated measurements in Fiji (not part of the primary workflow)

If specimens exhibit a simple, elongated body structure, are not curled or bent and lack long lateral appendages, and do not have antennae and cerci, automated measurement may be feasible. For such individuals, using Analyze > Analyze Particles can automatically extract size and shape data, including a value that approximates body length, under the right settings.

If Fit Ellipse is selected under Set Measurements, Fiji will give a parameter labeled “Major”, representing the major axis length of the fitted ellipse. This value can serve as a proxy for body length in species with relatively uniform, streamlined body shapes (e.g., many Diptera larvae).

Caution: The Major axis value might include parts - like antennae or cerci -, that are normally excluded from measurements. Furthermore, these appendages are often lost or damaged during sample preservation, and their presence or absence can distort measurement results. Therefore, for taxa with complex morphologies or missing structures, the Major axis value may not accurately reflect true body length and should not be relied upon as a substitute for manual measurement.

Note: Always validate automated measurements against manual ones for accuracy. See Fig. 22 for comparisons.

Alt-text: Unlabelled box dummy alt text

Fig. 22.

(A) Major axis measurements obtained through automated analysis - see Major values; (B) Manually measured body length values - see Length values. A comparison between the automated (A) and manual (B) measurements exhibits strong similarity for the first specimen, which lacks notable appendages. In contrast, the second specimen shows a significant difference due to its extended legs, leading to inaccuracies in the automated Major axis measurement.

Output summary - prior to the final step - each processed image should result in:

• •A labeled image showing each specimen.
• •A result table with individual morphometric data or an Excel file with the data copied in.
• •Manually measured body lengths in real-world units (added to the Excel file or in .csv format).
• •Optional: A saved list of ROIs (segmented outlines) for further review or re-measurement.

Body length estimation from surface area

This final step outlines how to estimate body length for specimens that were not measured manually, using taxon-specific regression models. Given the large number of samples and the high density of specimens within each bulk sample, manual body length measurement for all individuals is not always feasible. Instead, two-dimensional body area extracted from segmented images can serve as a proxy. By applying taxon-specific allometric regression models, accurate body length estimates can be obtained based on the empirical relationship between surface area and body length. To apply, follow these steps:

• 1. Prepare the dataset (for example, see Supplementary material 1)

In Excel or similar software, compile a table for each taxon containing:

• •Sample identifier (e.g., SampleID_SamplingDate) - if necessary for the study.
• •Taxon name.
• •Measured surface area (mm²) from Fiji.
• •Manually measured body length (mm) for a subset (∼5–20% of individuals per taxon; typically 50–100 specimens, depending on sample size and diversity), which serves as the basis for regression model construction.
• •Optional: other metadata (e.g., perimeter, major axis).

• 2. Build log10-transformed regression model in Excel

To linearize the power-law relationship between measured body length and the corresponding extracted surface area, both variables must be log10-transformed:

• •Create two new columns in Excel and calculate log10(Area) and log10(Length) using the formula: =log10(cell_with_value).
• •To visualize the relationship, insert a scatter plot with log10(Area) on the X-axis and log10(Length) on the Y-axis (Fig. 23).
• •Use the built-in Excel functions =SLOPE() and =INTERCEPT() to determine the regression coefficients more accurately by selecting the appropriate cell ranges containing log10(Area) and log10(Length) values.

• 3. Back-transform the regression equation

Fig. 23.

The relationship between log10-transformed body surface area (x-axis) and log10-transformed body length (y-axis) illustrated on linear regression plot. Each point represents an individual specimen. The fitted regression is shown with its corresponding equation and coefficient of determination (R²), which indicates the strength and explanatory power of the model in describing the relationship between the two variables.

Once the slope (a) and intercept (b) are obtained, apply the back-transformed formula to estimate body length from area:

Length = 10 ^{(a*log10(Area)+b)}

This formula can be applied in Excel:

=10^((a*log10(cell_with_area))+b)

• 4. Apply the model to unmeasured individuals

Use the taxon-specific equation to estimate the body length for all individuals of the same taxon that were not manually measured.

• 5. Model selection and quality control

• •Construct separate regression models for each taxon.
• •Include individuals from different sites or sampling dates, if possible.
• •Validate model fit using the R² value:

•     - Acceptable fit: R2 ≥ 0.75.

•     - Ideal fit: R2 ≥ 0.85.

• •If model fit is poor, check for outliers or measurement errors.
• •If a reliable model cannot be established, consider manually measuring the remaining individuals.

Taxon-specific regressions were used because body shape, proportions, and scaling relationships between area and length vary substantially among taxa, making pooled models less accurate and potentially biased.

Note:

• • The logarithmic transformation was used to linearize the non-linear relationship between area and length, improving the accuracy of the regression model and ensuring homoscedasticity of residuals.
• • The back-transformation using base-10 exponentiation yields estimates in the original measurement units.

How to use this method may depend on the research objectives and the precision required. In some studies - such as those using composite bulk samples from a single site - measuring 50–100 individuals per taxon may be sufficient to estimate average body lengths. This range was found to provide stable regression relationships (consistently high R² values) across taxa in our empirical tests. For broader-scale studies involving multiple sampling sites, dates or countries, the same number of measurements per taxon per location / per date can provide representative estimates.

To facilitate this process even more, a full R script has been developed (see Supplementary material 2), which builds linear regression equation per taxa and calculates the body length values for the unmeasured specimens using the derived regression models - provided the models meet a minimum reliability criterion of R² ≥ 0.75.

Method validation

Regression-based body length estimation from automated image segmentation showed strong agreement with manual measurements (R² = 0.817), even when evaluated at a higher taxonomic scale using the six most numerous and frequent groups in bulk samples (see Fig. 24). The validation dataset included approximately 1400 individuals across multiple taxonomic groups, covering a broad range of body sizes and morphologies.

Fig. 24.

Manual vs. Estimated Body Lengths for six major macroinvertebrate groups (Coleoptera, Crustacea, Diptera, Ephemeroptera, Plecoptera, Trichoptera). Each point represents an individual specimen. (A) On the scatter plot, each taxonomic group is coded by a different color. The dashed line indicates the 1:1 relationship. The majority of data points cluster close to the 1:1 line, demonstrating strong agreement between manual measurements and estimates derived from taxon-specific allometric regressions. (B) In the Bland-Altman plot, the y-axis shows the measurement difference (manual-estimated) against the mean of the two measurements (x-axis). The blue line indicates the mean bias, and the dashed red line represents the 95% limits of agreement. The plot shows that differences are evenly distributed around zero, with no strong size-dependent bias, indicating good agreement between the two methods.

In general, taxon-specific regressions demonstrated strong fits (R² ≥ 0.85 for most taxa). Some taxa of Coleoptera and Diptera exhibited slightly lower accuracy, likely due to e.g., morphological variability, but these deviations remained within acceptable limits, confirming the model’s overall reliability. Accordingly, taxa with higher morphological variability may require increased manual validation of segmentation and regression outputs compared to more morphologically uniform groups. The combination of high R² values (typically ≥ 0.85) and agreement patterns observed in Bland-Altman analysis indicates that the regression-based estimates provide sufficient accuracy for ecological applications. The estimation errors were small relative to specimen size, with no evidence of major systematic bias, supporting the method’s suitability for reproducible, high-throughput morphometric analysis. In this context, finer taxonomic resolution may further improve model performance by reducing within-group morphological variability.

Limitations

Although the method described here was designed to enhance throughput, improve reproducibility, and expedite image analysis, it has certain limitations that need to be mentioned.

First, the lack of high-contrast image or having mixed taxa with suboptimal background, can slow down the image processing workflow and can also affect the image segmentation accuracy. These factors also limit the transferability of trained classifiers in Labkit, as classifiers may not perform consistently across images taken under varying conditions.

In addition, the regression-based estimation of body length relies on the relationship between two-dimensional surface area and one-dimensional body length. This relationship may vary depending on body shape, posture, or life stage of the specimens. Specimens that are curved or not fully flattened may introduce minor bias into length estimates. The use of taxon-specific regressions helps mitigate this issue, as regression equations reflect typical shapes and size ranges for each taxon, and specimens that could introduce bias due to posture were excluded from regression modeling. Validation was performed on the most abundant taxa, which represent a wide range of body sizes and morphologies typical of freshwater macroinvertebrate assemblages; however, performance for rare or highly specialized taxa may vary.

Second, we faced taxonomic limitations. For example, Oligochaeta were excluded from analysis due to their fragmented appearance in images, which made it impossible to extract valid area or length measurements. More generally, certain taxa, such as soft-bodied organisms, coiled larvae, or fragmented specimens, may not be reliably measured using automated image-based approaches. When necessary for dataset completeness (e.g., rare taxa), these specimens can still be measured manually from the original images.

Third, while several steps are semi-automated, manual intervention remains necessary - for example, during segmentation correction or manual length measurement. This may introduce some operator dependence but can be minimized through clearly defined criteria and consistent workflow application. In particular, overlapping, or clustered specimens may reduce segmentation accuracy, and careful manual correction is recommended to ensure reliable measurements. Although such corrections may be required in high-density samples, they are typically localized to clustered individuals and affect only a small proportion of specimens, remaining manageable without substantially limiting overall workflow efficiency.

Finally, in our research the goal was to get genus-level body length dataset, but the images were taken based on broader taxonomic groups. So, we had to include an additional and time-consuming step, the re-identification (for details see Supplementary material 3) prior to manual length measurements. To avoid such workflow inefficiencies, we recommend that, depending on the taxonomic resolution required by the study, only individuals belonging to the relevant taxonomic group be included within a single image frame during specimen photography.

These limitations should be considered when applying the workflow, particularly in relation to sample composition, imaging conditions, and taxonomic variability. With careful imaging practices and consistent protocols, the method remains robust and widely applicable.

Ethics statements

The authors have read the ethical requirements for publication in MethodsX and confirmed that this study did not involve human participants, animal experimentation, or data collected from social media platforms.

CRediT author statement

Anita Szloboda: Conceptualization, Methodology, Investigation, Writing - Original Draft, Visualization, Project Administration Zsolt Kovács: Investigation, Writing - Review & Editing, Visualization José Barquín: Conceptualization, Writing - Review & Editing Gabriel Singer: Conceptualization, Methodology, Writing - Review & Editing, Supervision, Project Administration, Funding acquisition Zoltán Csabai: Conceptualization, Writing - Review & Editing, Supervision, Project Administration, Funding acquisition

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used ChatGPT (OpenAI, 2025) in order to support the writing and development of this manuscript, to assist with drafting, improving the readability and generating example R scripts. After using the OpenAI tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Data will be made available on request.