Frequently
asked questions

VAIDR analyzes cells. To do this well, it combines several powerful capabilities into one package: 

• The AI is trained to the specific requirements of your application. This ensures that the AI will actually work on the myriad of diverse applications in the field. 
• To make this practical, VAIDR has an intuitive, smooth workflow. It is easy to learn, time-efficient & fun to use. 
• Seamless integration. From samples to the final results, you can do everything within the VAIDR platform. 

VAIDR offers very powerful image analysis for very little effort.

Whatever you can consistently see in your images, VAIDR can quantify. Often, even things you do not see. Due to the use of AI-algorithms, amazing analysis results can be achieved even on label-free images.

In contrast to other techniques, it does not take special skills or frustrating efforts to get there. VAIDR makes the use of AI in image analysis practical.
 

VAIDR can work with all cell types. Adherent, suspension, aggregates, organoids – you name it. The algorithm is not limited to any particular application. For any given set of images, VAIDR’s user-trainable AI automatically constructs the proper analysis algorithm to answer your research question.

Because VAIDR uses user-trainable AI, the possibilities are endless. But here are some examples: Viability, cell count, substructures like nuclei, nucleoli, organelles, neurites. Co-culture analysis, unbiased scoring of treatment effects, detection of genetic defects, quantification of differentiation phenotypes, ...

Have a couple of nice reads with our application notes or sit back and watch some videos on our YouTube channel.

There are three main steps: Data input, training, and analysis. 

1. For data input, you either upload images from your existing microscope, or you use the VAIDR microscope, which is directly integrated with the platform. 
2. For training, you either use our labeling editor in the segmentation workflow to teach VAIDR what you want to detect and quantify. In the classification workflow, you simply assign entire groups of wells to your classes of interest and train a validated classifier. 
3. For analysis, you simply apply the trained methods to lots of images. You can either evaluate the quantitative results inside the VAIDR platform or export them.
    

The VAIDR microscope provides digital phase contrast and one fluorescence channel of your choice. The microscope automatically images any type of plate or cell culture dish using auto-focusing. The imaging workflow is highly automated – no need to tune any settings.

The VAIDR microscope makes high-contrast label-free images which are great for machine-learning. In many cases, the contrast and level of detail are better than in standard phase-contrast microscopes. The additional fluorescence channel extends the range of applications significantly.

The VAIDR microscope is seamlessly integrated into the platform. This means you do not need to manage your images anywhere else and then upload them, as is the case when using a third-party microscope.  

So, if you do not currently have an automated imager and the imaging modalities of the VAIDR microscope are sufficient for your use-case, it is a great option.
 

Yes, definitely. VAIDR has a simple upload interface where you can upload TIF images for entire plates including plate layout data. Since almost all microscope software packages have a TIF export option, you can use virtually any microscope as a data source for VAIDR.

For the segmentation workflow: 

It depends mostly on how variable your phenotypes are. In many cases, a surprisingly accurate segmentation can be trained by labeling a single image. Most of the time, segmentation models require less than 10 labeled images. Due to the smart labeling workflow, most of these images do not need to be labeled completely, but the user only needs to correct the mistakes that an earlier method version made, to train an improved version.

For the classification workflow: 

Since this has a built-in statistical validation feature, the data requirements depend on the level of validation you need. For a quick check whether phenotypes are distinguishable, images from a few replicate wells per condition are sufficient. To establish a routine classification workflow, images from several batches/timepoints etc. may be required.
 

This ranges from 1-2 hours for a small segmentation training to several hours for a large study with a comprehensive validation package.

Things to consider are: How do you intend to validate the trained method and against what type of variability do you need your method to be robust? In general, the training data should encompass all the variability that the trained method will encounter during its application.

So, for a one-off demonstration, within a fixed data set, there just need to be sufficient test data that were not used for training, to demonstrate that the method has learned something relevant and did not just memorize the training images. This principle applies generally: Always set some test data aside to check if the method generalized from the training.

On the other end of the spectrum, if the method is intended to analyze data that will be generated in the future, it is important to keep an eye on the performance and potentially add new training data to update the method. This may be necessary when the encountered phenotypes evolve over time.

A better answer to this question usually requires more specific knowledge about the application. The VAIDR team are always available to help with experimental design and best practices to make your project succeed.
 

First, what are the alternatives? 

The most common analysis method is surely the expert’s eye. An expert grading 1000 images will see a lot of detail in them---but they won't look at the 1000'th image the same way they looked at the first. Especially if the first is right after lunch and the last is just before end-of-shift. 

An expert assessment based on a large image set is therefore highly variable. 

What's more, the expert's time is utterly wasted on that task; they should be working on optimal culture conditions and effective experiments etc., not on grading images. 

At the other end of the spectrum, images are evaluated using simple intensity means or pixels are counted with an intensity above a user-defined threshold. This method does have the advantage of being quick and simple, both to develop and to apply. However, it is strongly affected by differences between samples in staining, illumination, cell density and many other factors. What’s more, only a small fraction of interesting scientific questions can be answered by such simplistic read-outs. 

In between are hand-written image analysis scripts. They do produce consistent results on bulk data, and developers can tune them to detect very subtle effects. Sometimes, scripts perform better than AI machine learning algorithms because the experts writing them can pour all their background knowledge into them. 

However, the kind of talent that produces scripts that perform as well as or better than AI is rare and should be applied only to the most valuable and most difficult tasks. 

AI, on the other hand, can learn to detect very subtle effects robustly and at scale, and all it needs is data and (biological) expert input.
 

VAIDR uses state-of-the-art AI algorithms to analyze your data, i.e., it will perform about as well as currently possible on the input. 

What really sets VAIDR apart, however, is the way the AI integrates into the user’s work:

1. We believe that your analysis should be the exact analysis you need for your application. You shouldn’t have to buy whatever analysis package we’ve happened to develop that is closest to what you really need.
2. Therefore, you should be able to use your own data and expert knowledge to develop exactly what you need yourself. And it should be easy and quick. No coding or AI skills should be necessary.
3. We realize that scored images have no value for you. The value is in assay results for your samples, conditions, experiments. So, VAIDR supports the entire workflow from acquiring your data, through managing it, developing and applying morphological read-outs, all the way to tracking, plotting, and statistical tests as needed to get your results.
    

AI or image analysis skills are not needed to use VAIDR. If you know how to use spreadsheets and if you’ve ever used a simple drawing program on your computer, you should be fine.

VAIDR learns from you. Depending on the task, you either teach VAIDR to find known target structures and quantify their area or shape, or you teach it to assess whole images and score them along the criteria that you care about:

In the first case, you use VAIDR’s inbuilt Labeling UI, which is a specialized drawing program for coloring the structures you are interested in. VAIDR then simply learns to do what you do. 

In the second case, you simply use VAIDR’s data filtering tools to give VAIDR a few examples of data from different conditions, like 'sick'/‘healthy’, 'treated’/'untreated’, ‘QC OK’/‘QC failed’ etc., and VAIDR learns to distinguish images from those conditions. Either way, VAIDR can then apply what it’s learned to more images, quantify the results and help you analyze them in the context of your experiments or production processes.
 

VAIDR emphasizes transparency and rigorous methods. It includes sophisticated and stringent method validation procedures, and it allows the user to drill down from validation reports and aggregated statistics and graphs, through scores of individual images, and finally target structures like cells or parts of cells.

Every time you image a sample or upload images, VAIDR asks you to enter metadata, like cell identity, treatments, culture conditions, or whatever else is relevant to your application. VAIDR uses this information to keep a digital lab book.

You can then search your data by the metadata you have. VAIDR provides you with friendly tabular and graphic views on your metadata and allows you to quickly find and inspect any images you’ve put into the system.

Just like the primary imaging data, analysis results are tied to the metadata you enter when you feed images into the system. You can use that metadata to navigate down to the images and even the pixels and individual objects, like colonies, cells, nuclei or whatever else you’ve taught VAIDR to recognize, and seen quantification results for each.

Or, you can aggregate quantifications per image, or well, and see the aggregates in the electronic lab book, next to the metadata, or plot them, grouped according to the metadata, and perform statistic analyses.

After you’ve imaged samples or uploaded your data, developed your morphological assay, and applied it to your experiment’s data set, you can use VAIDR’s built-in, intuitive plotting and statistics functionality to plot the analysis outcomes across your experimental dimensions. You can also ask VAIDR to compute statistical tests to answer your experimental questions.

Additionally, you can download either the aggregated data underlying the plots and statistics or the raw morphological read-outs for further, customized processing on your part.
 

VAIDR is often used in production processes, where morphological quality metrics are developed and standardized to optimize production or keep quality stable.  Within minutes after imaging, VAIDR can give you these metrics, visualize them, and let you compare them with historical data. 

Within the VAIDR application, you can see when your quality fluctuates and you can investigate which parameters drive fluctuations. You can perform lot tests of your reagents and you can use VAIDR for your release assays.
 

VAIDR is an open system. At any point, you can download images, metadata, and analysis results in standard file formats that you can use for downstream processes or documentation. 

VAIDR also provides an open REST API, which makes it easy to integrate it with external systems which need automatic, seamless data transfer.
 

Yes, it can. VAIDR offers a REST API that can be accessed to control the system and query and transfer data. An automation platform or LIMS can use this REST API.

Alternatively, we can offer customization services at very reasonable pricing to integrate VAIDR into your infrastructure—either in collaboration with your infrastructure manufacturer or on our own.

Yes, absolutely. Like any instrument or method, AI analyses produce read-outs for your samples. If you pay attention to the usual methodological niceties—control for confounders, avoid biases, don’t fit and apply on the same data…—then any effects you see are likely real.

As always, it is necessary to cite all the materials and methods you used. We’ll be happy to help you with this in the case of our VAIDR system.

It all belongs to you. We do not claim ownership of the image data, the metadata, the analyses you develop, or the analysis results.

Yes, it is possible to transfer analyses between VAIDR systems and online accounts. This is a good way to ensure comparability between sites and projects, and to save time, effort, and materials.

TRI offers different VAIDR configurations for VAIDR Online users. Customers can choose between license types varying in upload capacity and duration. VAIDR Online users can expand their imaging capacity by a cloud-connected automated microscope to integrate the image data workflow. A full VAIDR System is the most powerful tool in the VAIDR portfolio: A full VAIDR shows the highest performance and no data limits for high volume users or multi-project labs using VAIDR for reproducible experiments and robust QC cell analysis.

Not sure what’s the best VAIDR solution? Click your configuration here.

Ideas might grow while working with VAIDR, and VAIDR can grow with your requirements. You can change hard- and software licenses at least with the next leasing contract, e.g. If you need a second microscope or you might plan to change the hardware configuration. Within the leasing period upgrades are possible.

Please ask for a quote, if you are interested.

We want as many satisfied customers as possible in the VAIDR community and therefore, we offer a complimentary trial phase to our new customers. If you have an application in mind to be solved by VAIDR, please have a look at our published Application Notes or contact Thomas Frahm. 

Depending on your VAIDR configuration, there are opportunities to lease or to buy VAIDR. VAIDR Online only comes with a quarterly or annual license. Besides annual leasing the hardware can be purchased. Please ask for a quote, if you are interested.

If you order VAIDR as a web solution, cloud hardware or an instant full VAIDR system, all configurations need internet access with adequate stability, bandwidth (minimum 5Mbit/s upstream and downstream) and security to facilitate the provision of the owed services.

VAIDR needs the internet connection for the ordered performance, the provided online support and system services.
 

Using VAIDR is easy to learn. Our personal onboarding training ensures that you can use the entire system's functionality. Additionally, TRIs support team offers continuous online support to VAIDR users when needed. VAIDR comes with an included service & support package throughout the license term.

Are you offering any support in using VAIDR and/or to set up experiments?

Yes, personal & interactive support is available by telephone, email and chat (Teams, Signal). There is supportive information on TRIs YouTube-Channel and Application Notes on VAIDRs webpage. The material is constantly updated with new stories, tricks and procedures to keep the user up to date.