Search
AI image enhancement and segmentation software for materials characterization and quality control
Thermo Scientific Avizo Software’s AI tools enhance and segment 3D electron microscopy and CT datasets, enabling faster and more accurate feature retrieval compared to traditional methods.
When integrated with Avizo Software's trusted visualization and analysis environment, you gain efficient, repeatable workflows that speed time-to-results.
How Avizo Software’s AI capabilities improve materials analysis workflows
Avizo Software’s AI tools simplify and accelerate imaging workflows, delivering deeper insights so you can concentrate on scientific interpretation rather than software operations.
Faster discovery
AI routines streamline processing and cut long tasks to minutes, delivering precise results for faster discovery.
Easier accessibility
No specialized knowledge required. With intuitive AI tools, your answers are easily accessible.
Shareable models
Leverage pretrained AI models or share your own model with your colleagues and peers.
AI-enhanced image analysis in Avizo Software
- Raw Data
- Labeling
- Training
- Inferencing
- Processed Data
- Report
- Pre-trained Model
Get insights from raw 2D and 3D data with AI segmentation
Avizo Software natively accepts most 2D and 3D imaging datasets such as FIB-SEM, TEM, 3D spectroscopy, and micro-CT. It is often useful to view the data in multiple angles and in 3D to qualitatively assess image quality and determine the best approach for analysis.
2D or 3D microscope data may not consistently delimit features important for quantitative research.
For example:
- Metal artifacts in micro-CT
- Charging artifacts in SEM
- Pore back in FIB-SEM tomography
- Missing wedge in TEM tomography
These datasets are amenable to AI segmentation, as this can be more efficient compared to traditional methods.
3D FIB-SEM tomography of a geological sample with a cross-section inset as viewed in Avizo 3D Software. Note the uneven charging of this insulating sample. Though your eyes can track features, they would be hard to segment with traditional methods; thus, this is a good candidate for AI segmentation. Data collected on a Thermo Scientific Scios 2 FIB-SEM.
Reduce labeling time with AI-assisted selection tools
Avizo Software provides several flexible and efficient ways to generate ground truth. A dedicated interface, called the Segmentation+ Workroom, lets you choose regions of interest (ROIs) for training and to label data, train models, and perform model inference all in one interface.
- Selection methods include manual drawing and automated selection tools such as Brush, Lasso, Magic Wand, and Thresholding.
- AI-Assisted Selection can be used as a pre-trained model for quickly selecting visually distinct objects.
Avizo Software also accepts labels from other methods such as Image Recipes or from external tools.
The Segmentation+ Workroom within Avizo 3D Pro Software. A lasso tool is used to select the carbon binder containing Pt nanoparticles within a rectangular region of interest. Only this ROI is used to train an AI model. The data are from TEM tomography, so the borders are fuzzy and indistinct in some areas. This means AI segmentation is the best approach for these data. TEM tomography data collected on a Thermo Scientific F200X G2 (S)TEM.
Train custom AI models
Avizo Software equips you with a complete system to train custom AI models.
- Denoising models receive one or two inputs and produce a re-usable model.
- Class Segmentation models are provided with two inputs for distinguishing materials, such as composite constituents or porosity.
- Instance Segmentation models distinguish individual objects from one another and from the background.
You can prioritize fast, lightweight models for initial segmentation or higher precision using deeper networks. The interface prioritizes ease of use with advanced options available to experienced users.
In-progress and completed model training can be interactively monitored in a browser. This lets you compare precision and accuracy metrics between models and visualize model progress.
AI model training progress is visually tracked through a browser using TensorBoard. This is launched automatically from Avizo Software when training an AI model. On the left, raw data are shown. The center column shows ground truth in red against a black background. The right column shows model prediction, and an interactive slider lets you see how the model refines and improves with training time. TEM tomography data collected on a Thermo Scientific F200X G2 (S)TEM.
Choose the right AI model for your data
To assist in selecting the best custom or pre-trained AI model for each dataset, Avizo Software includes an AI Model Evaluation tool. Here, one or more AI models predict a subset of a dataset, and results are shown with standard metrics and an interactive visual preview.
Upon selecting a model, prediction can be run. This will yield either segmentation or denoised data, depending on the choice of model. Results are immediately ready for analysis.
When you have more than one eligible AI model for a dataset, it is useful to test which is the best fit for new data. In this AI Model Evaluation interface, metrics about precision, accuracy, etc. are shown and a visual result is shown on an ROI. Once a model seems best, the model is selected and used for prediction. This image shows results from micro-CT of porous copper foam with cracks in the copper.
DOI: 10.1016/j.matdes.2013.07.047.
Analyze AI-processed data
Ultimately, AI methods represent an intermediate step for extracting quantitative information from imaging data. Avizo Software includes many measurement tools based on segmentation such as percent volume fraction, per-object measures, interface analysis, and connectivity.
Specialized workflows include:
- Triple Phase Boundary measurement, such as for SOFCs
- Pore Network Modeling (XPNM)
- Fluid flow and ion diffusion simulation (XLab)
- Volumetric mesh generation for FEA and CFD (XWind)
- Displaying Digital Volume Correlation results (XDVC)
- Efficient fiber phase extraction from complex materials for individual fiber analysis (XFiber)
Explore downloadable techniques and workflows for use with AI model outputs in the Xtra Library.
A composite image showing 3D FIB-SEM on the left, pore phases segmentation in the center, and a pore network model on the right. This method uses the XPNM extension in Avizo 3D Pro Software on AI model segmentation results. DOI: 10.5281/zenodo.4049960. Data collected on a Thermo Scientific Helios DualBeam.
Generate templated reports
For traceability and quickly delivering information to decision makers, generate templated reports based off of your work in Avizo Software. Though not restricted to AI-empowered work, the Reporting function can add an additional level of trust to what AI models are doing.
Reports include statistics, charts, images, metadata, and custom fields for documenting the work done in Avizo Software for those who are not using the software. PDF export ensures that the work is easy to share and act upon.
Report generated by Avizo Software for a composite fiber material used in automotive applications. Standard reports like this show the process of an Avizo Software project, with or without AI methods, and reports statistics based on the report author's choice.
Pre-trained AI models accelerate image analysis
It is not always necessary to train your own AI model. Re-using an existing model saves considerable time while still giving you high-quality outputs from AI methods.
- In Avizo Software it is simple to re-use AI models you or colleagues have previously trained on similar data.
- Use the AI Model Evaluation tool to decide which model best suits your data.
- Explore pre-trained models available for download in the Xtra Library, such as CellPose.
- Consult with our team to generate AI models to suit your application.
3D electron backscatter diffraction (EBSD) volume overlaid atop corresponding 2D SEM cross-section from FIB-SEM tomography data. Here, the indexed EBSD data was used as an input for a pretrained 3D CellPose instance segmentation model in Avizo 3D Pro Software. The result is a per-grain segmentation for shape and orientation characterization. Data collected on a Thermo Scientific Helios 5 PFIB-SEM.
See Avizo Software's AI tools in action
Discover examples of AI workflows using Avizo Software in our on-demand webinar.
Characterization of SOFC anode using AI
Improve design to increase performance and longevity
AI-boosted crystallography for high-end alloys
Increase material resistance to fracturing due to physical stress
AI detection to improve nanomaterials analysis
Nanoparticles engineering for next gen materials performance
Predictive AI model for materials composition
Improve corrosion resistance of coatings
Avizo AI imaging software FAQ
Try it, you might like it. AI methods are relatively new and can work for noisy or non-ideal images that traditional image processing methods may not handle well. Avizo Software includes both AI methods and image processing methods. Sometimes, the combination of these techniques enables otherwise difficult work to be done more efficiently.
It depends. Reusing and training 2D segmentation is included with Avizo 3D Software, as is training and using AI-based denoising. Advanced features such as the AI Assisted Selection, Cellpose, and training 3D deep learning models require an Avizo 3D Pro Software license.
No. You do not need prior programming or machine learning experience. The AI functions in Avizo Software are integrated into the UI as a guided workflow for typical use or as modules for integration into Recipes for repetitive tasks.
Not much. In many cases, a 2D model is rather effective for segmentation tasks. Training a 2D model can usually be accomplished by labeling 2–5 regions of interest. If the data are more variable or heterogeneous, you can train a shallow network to assist with labeling additional regions of interest, effectively using AI models to more efficiently train AI models.
Not long. In many cases, models train in 30 to 60 minutes. This depends on data size, hardware, and choice of model parameters. Typically, prediction is relatively fast and can be nearly instant to a few minutes for very large datasets.
Yes. Avizo 3D Pro Software includes the AI Assisted Selection tool, which is a pre-trained AI model. Other pre-trained models are available for download using a built-in Xtras Installer. You can browse these tools in the Xtra Library.
Yes. AI methods are quickly being embraced in many image-based research disciplines, including peer-reviewed research and production environments. Once trained, AI models (as we implement them), are consistent and can be included as intermediate outcomes in the Reporting Workroom. Avizo Software makes it simple to retrieve the model type, training data, and evaluation metrics that are typically requested to ensure a model is reproducible and effective.
Yes. Avizo Software includes a dedicated model import function for using models within the Segmentation+ Workroom. Also, previously trained models can be used in the Project Workroom much of the time. In some cases, there are important differences between software versions, so please contact us if you need support transferring AI models between software versions.
It is required to have an NVIDIA GPU with CUDA support. There are other considerations shared with the rest of Avizo Software. As these requirements may change, see our full list of hardware requirements.
We offer training sessions, video tutorials, webinars, and technical support to ensure you are using Avizo Software to its fullest capacity. It is recommended to begin with the User’s Guide packaged with the software or our Learning Center.
Experience Avizo Software
Are you struggling to make sense of complex data sets? Extracting meaningful insights can be a daunting task. Avizo Software is here to help.
With its advanced visualization and analysis capabilities, Avizo Software turns your data into actionable insights, empowering you to make informed decisions.
Ready to experience the difference?
Request a trial today and see how Avizo Software can transform your data analysis.
For Research Use Only. Not for use in diagnostic procedures.