Joint Webinar with Dragonfly
Advanced 3D CT Workflows for Complex Structures
This is a written summary of a live webinar presented on October 22, 2025. The recording and resources are available on the recording page.
Presented by:
Webinar summary
This webinar, co-hosted by Rigaku and Dragonfly, walks through a practical, end-to-end CT workflow using a concrete core as the running example and keeps the message clear: useful results come from balancing physics limits, good data collection, and fit-for-purpose analysis. On the hardware side, Rigaku’s benchtop CT systems are presented as fast, flexible tools for lab work, with scan times that can be seconds, maximum tube voltage around 130 kV, and geometric setups that let you trade field of view for resolution by moving the detector and part. The presenters emphasize realities experienced users know well but don’t always hear said out loud: big, dense parts will limit X-ray penetration; pushing magnification on large parts costs you resolution; and artifacts are a fact of life. They show the set-up details that matter in practice—using rigid or 3D-printed holders to keep irregular specimens stable, centering by rotating the sample before scanning, and picking basic parameters by short test scans. For the concrete sample, settings in the ballpark of 110–130 kV with modest current and an aluminum filter produced workable contrast, and the operator leaned on Rigaku’s database pane to reuse and compare scan recipes. Reconstruction is framed as an equally critical stage rather than a button click: the team discusses algorithm choice, beam-hardening correction (and why the filter sometimes lets you skip it), ring-artifact correction, and denoising, all compared side-by-side to decide what actually improves the volume rather than just smoothing it.
On the analysis side, Dragonfly is positioned as the post-processing workhorse for people who need to quantify rather than just visualize. The demonstration starts with a “one-click” porosity segmentation that excludes outside air and reports pore volumes immediately, then steps up to more advanced jobs like crack mapping with voxel- and sub-voxel-based thickness measurements, and pore-network modeling with tortuosity. The message is pragmatic: thresholds and presets get you fast, defensible numbers on simple, single-material data; once materials or artifacts complicate things, you graduate to deep learning. The live training example is intentionally small and realistic—three classes (air, material, pores), a quick round of painting to teach the model, and a few epochs to a result that is more robust than thresholding on noisy volumes. Trained models are saved to a library for repeatable batches, which is exactly what most industrial users need when they’re processing series of plugs, AM test cubes, or routine QA parts. For those who need downstream CAD or metrology, the latest Dragonfly adds sub-voxel surface determination and clean STL export, and the Q&A confirms common pain points: tortuosity is doable but compute-heavy on large data, voxel size and scale flow automatically from Rigaku’s VOX metadata so you’re not guessing units, and there are add-ins and presets for specialized domains like bone microstructure. The closing note is straightforward and not salesy: if you’re new to CT or wrestling with tricky samples, you can bring parts into the Rigaku lab and run through this exact workflow with help; if you already scan but struggle to get numbers you trust, the Rigaku-Dragonfly pairing is designed to get you from a decent scan to quantitative results quickly, with enough headroom to tackle harder, multi-material datasets when you need it.
Key questions answered in the webinar
-
If the part is too big or too X-ray-opaque, the tube power simply won’t push enough photons through, so you must balance part size and material attenuation against the scanner’s capability rather than forcing a “high-res” recipe that physics won’t support. Magnification is the other hard constraint: bring a small part closer to the source and you gain detail; push a large part farther away and you lose it. On the benchtop systems shown, the source is fixed while the detector and stage positions let you trade field of view, resolution, and penetration; for oversize specimens you can use offset scanning and stitched heights. The specific platform demonstrated tops out at 130 kV and can run very fast scans, which helps with throughput but doesn’t override those geometric and material limits.
-
Finding the right or optimal CT settings to achieve the best combination of performance, resolution, and scan time can be challenging, even for experienced operators. The key factors include:
- X-ray penetration and absorption: X-ray radiation passes through the sample, and different component materials and phases absorb the radiation differently (e.g., iron absorbs relatively large amounts, carbon absorbs significantly less). The operator must find a balance between the size of the sample and its absorbing properties on one hand, and the power of the X-ray computer tomography scanner on the other. If a sample is too large or dense, the X-ray tube power may not be efficient.
- Magnification factor and resolution: The magnification factor is vital for obtaining the highest possible resolution and resolving the smallest details of the sample. Placing a small sample closer to the X-ray source improves resolution, while a large sample must be placed further away, resulting in somewhat worse resolution.
- Software parameters: The control panel allows the operator to vary many parameters, including changing the voltage, current (e.g., 110-130 KV and 120-150 microampere were used for the concrete sample), and focal spot size of the X-ray source. Other crucial settings involve selecting a mechanical filter (e.g., aluminum or copper), correcting the field of view, adjusting detector properties (like pixel size/bin), and determining the number of two-dimensional transmission projections and iterations.
-
After the measurement is complete, Rigaku's software signals success and allows the user to select settings for 3D reconstruction. This involves several critical steps:
- Algorithm selection: Various mathematical algorithms for 3D reconstruction are available, allowing the CT operator to find and compare the optimal one for their specific sample.
- Beam hardening correction: This is important for many applications, particularly dense samples like concrete. Since X-rays are polychromatic and different energies are absorbed differently, the software implements tools to compensate for this effect by entering correction values.
- Post-processing corrections: The software also offers other options, such as additional mathematical denoising of CT images and ring artifact correction. For the specific concrete sample discussed, the operator used ring artifact correction and noise reduction options, but skipped beam hardening correction due to the use of an aluminum filter.
Once reconstruction is complete, the software displays the 3D reconstruction, along with three projections (X set, X I, and I set) on corresponding virtual two-dimensional slices. The software panel on the left provides icons for post-processing options like zooming, rotating, changing the gray value to improve visualization, and changing virtual slices.
-
Dragonfly is an advanced software solution for 3D visualization, quantification, analysis, and segmentation of 2D images (e.g., from microscopy), CT images, or 4D time-lapses. Dragonfly is known for several core capabilities:
- Visualization: It provides really nice visualization of results, with simple presets making 3D rendering comfortable for starters.
- Segmentation tools: It offers a wide range of segmentation tools.
- Deep learning (AI): It incorporates state-of-the-art deep learning algorithms to make data segmentation easier.
- Image analysis: It is excellent in image analysis and includes integrated statistics and reporting capabilities, allowing quantitative data (like volume or surface area) to be exported as CSV files, text files, or histograms.
- Automation: It features Python automated tools for users interested in scripting.
Dragonfly automatically accesses necessary metadata, such as the voxel size, from Rigaku's .v or .dbox files, eliminating the need for manual input when data comes from a Rigaku system
-
The automated porosity tool runs a single-material Otsu-based segmentation that finds internal pores while excluding those connected to outside air. From there, Dragonfly builds a multi-ROI you can measure immediately—volume fractions, size stats, and more—and you can visualize results with a live 3D “splitter” view to show pore distributions in context. It’s meant for fast, defensible numbers on clean data.
-
Use thresholds for simple, single-material data when contrast is clean; switch to deep learning when artifacts, stitching, or multiple materials make thresholding unreliable. The segmentation wizard trains on a small framed region with just a few painted examples for three classes—air, material, pores—then runs short epochs and gives you a model you can reuse on whole parts and future batches. Training large volumes takes longer, but the payoff is repeatability across many similar parts.
-
Deep learning (DL) is used when simple thresholding is insufficient, particularly if the images are irritated by artifacts (e.g., from dense material) or noise, or if the user requires an automated version of segmentation. The process involves "teaching" the deep learning model:
- Defining classes: The user defines the classes they need to differentiate, such as air, material, and pores.
- Teaching the model: The user identifies these classes within a small frame of the data set using tools like thresholding or painting/masking. This initial manual definition serves as the training input.
- Training: The model is trained using epochs, applying the taught information to other frames. Although large data sets take longer, training a model once allows it to be applied repeatedly to hundreds of parts, making it beneficial for complex or repeatable samples (like plugs or drill cores).
- Publishing: Once training is complete and the best result is automatically captured, the model is published and integrated into the user's deep learning library, ready to be applied to the full 3D data set.
-
Dragonfly supports several methods for analyzing complex structures:
- Crack analysis: Dragonfly is ideal for crack analysis due to its voxel-based analysis, which makes inspection fast and intuitive. It allows measurement of thickness, volume, or surface area. It achieves high accuracy using sub-voxel thickness algorithms, which help overcome the resolution limitations imposed by larger voxel sizes in large samples. The results can be visualized, such as mapping the thickness of a crack inside a rock core using colorful visualization.
- Pore segmentation (Automated thresholding): For fast results, Dragonfly provides a one-click core segmentation solution, such as the single material basic Otsu method. This separates voxels into categories like matrix and pores, allowing immediate quantification of data like volume.
- Advanced segmentation (SLPO): The Segment Anything (SLPO) tool, based on a deep learning library from Meta, offers a one-click solution for quantifying structures like corns (sand) or rock pieces in both 2D and 3D. This approach also allows for the calculation of the pore network model and tortuosity analysis.
Subscribe to the X-ray CT Email Updates newsletter
Stay up to date with CT news and upcoming events and never miss an opportunity to learn new analysis techniques and improve your skills.
Contact Us
Whether you're interested in getting a quote, want a demo, need technical support, or simply have a question, we're here to help.