Automation of the Micronucleus Assay Using Imaging Flow Cytometry and Artificial Intelligence

Performing the micronucleus assay using imaging flow cytometry overcomes many limitations of traditional methods, including low throughput, score variability, and lack of visual confirmation of events. The main advantage of this technique is that all key events can be acquired using imaging flow cytometry and analyzed using artificial intelligence. This method has a potential to be used in large scale screening of chemicals and other compounds to test for toxicity at higher throughput than is currently available.

When building a new AI model, the main challenge is obtaining sufficient images of cells with micronuclei, so it is important to use the image tagging algorithms. To begin, launch the Artificial Intelligence or AI software. Under Experiment Type, click the radio button beside Train to start a training experiment for building the Convolutional Neural Network or CNN model, and click on Next.

On the Class Names, click on Add. In the pop-up window, type Mononucleated, and click on OK to add the mononucleated class to the list of class names. Repeat this process to add the other class names, such as Mononucleated with MN, Binucleated, Binucleated with MN, Polynucleated, and Irregular morphology.

Under Select Files, click on Add Files, and browse for the desired files to be added to the AI software, to build the ground truth data. Next on the Select Base Population screen, locate the Non-apoptotic population from the population hierarchy. Then right click on the Non-apoptotic population, choose Select All Matching Populations, and click on Next.

To assign a tagged truth population of mononucleated cells with micronucleus, click on the Mononucleated with MN class under Model Classes on the left, and then click on the appropriate tagged truth population on the right. Once all appropriate truth populations are assigned, click on Next. On the Select Channels screen, ensure that the appropriate channels for the experiment are chosen.

Here, choose Bright Field or BF as channel one, herds for DNA as channel seven, and then click Next. Finally, on the Confirmation screen, click on Create Experiment. Click on Tagging to launch the tagging tool interface.

Then click on the zoom tools to crop the images for easier viewing. And click on the slider bar to adjust the image size and decide the number of images to be shown in the gallery. Click on the Display Setting option and choose min-max, which provides the best contrast image for identifying all key events.

Next, click on Setup Gallery Display to change the color of the DNA image to yellow or white, which will improve the visualization of small objects. Click on Cluster to run the algorithm to group images with similar morphology together. Once clustering is complete, click on the individual clusters and begin assigning images to appropriate model classes.

After a minimum of 25 objects are assigned to each model class, the predict algorithm becomes available. Click on Predict. Once the predict algorithm has completed running, add objects from the predicted classes to the appropriate model classes.

Once a minimum of 100 objects are assigned to each model class, click on the Training tab at the top of the screen, and then click on the Train button. Once model training is complete, click on View Results to assess model accuracy. Once the model training is complete, click the menu button to define a new experiment.

Under Experiment Type, click the radio button beside Classify to start a classification experiment, and click Next. Click on the model to be used for classification, then click on Next. On the Select Files screen, click on Add Files.

Browse for the files to be classified by the CNN model, and then click on Next. Next, on the Select Base Population screen, click on the checkbox next to the Non-apoptotic population in one of the loaded files. Right click on the Non-apoptotic population, click on Select All Matching Populations to select this population from all loaded files, and then click Next.

On the Select Channels screen, verify channel one is selected for the Bright Field, and channel seven is selected for the DNA stain, and click Next. Finally, on the confirmation screen, click on Create Experiment. The AI software loads the selected model and all images from the chosen data files.

After loading, click on Finish. Next, click on Classify to launch the classification screen. And use the check boxes to choose to use Random Forest or RF, and CNN.

Then click on the Classify button. This begins the process of using the RF and CNN models to classify additional data and identify all objects that belong in the specified model classes. Once the classification is complete, click on View Results.

Click the Update DAFs button to bring up the update DAFs with classification results window, and then click on OK to update the DAF files. To generate the report, on the results screen, click on Generate Report. If an individual report for each input DAF is required, select the checkbox beside Create report for each input DAF.

Otherwise, just click on OK to obtain the reports. The AI assisted cluster algorithm groups similar objects within a segment together, according to the morphology of both unclassified objects and the objects that have been assigned to the ground truth model classes. Clusters containing mononucleated cells fall on one side of the object map, while the multinucleated cells are on the opposite side.

Binucleated cell clusters fall between mono and multinucleated cell clusters. Finally, clusters with irregular morphology, fall in different areas of the object map. The predict algorithm is more robust than the cluster algorithm in identifying subtle morphologies in images.

For example, Mononucleated Cells with MN versus Mononucleated cells without MN.The performance of the model can be assessed using tools, including class distribution histograms, accuracy statistics, and an interactive confusion matrix. In class distribution histograms, the closer the percentage values between the truth and predicted populations, the more accurate the model. In the accuracy statistics, The closer these metrics are to 100%the more accurate the model is at identifying events in the model classes.

Finally, the interactive confusion matrix indicates where the model missclassifies events. Genotoxicity was measured by the percentage of micronuclei by microscopy, indicated by clear bars, and AI indicated by dotted bars following a 3 hour exposure and 24 hours recovery, for Mannitol, Etoposide and Mitomycin C, using both the Cytochalasin B and Non-Cytochalasin B methods. When creating a training experiment, ensure the loaded data contains imagery from positive and negative control samples.

Micronuclei are rare, and sufficient imagery is required to build an accurate AI model. This procedure permits the creation of AI models to analyze imaging flow micronucleus data in any field of study, such as radiation biodosimetry. A rapid and robust AI-based method to identify micronuclei, may extend to other applications, such as quantifying micronuclei that may be predictive for the risk of cancer development.