Artificial Intelligence Approaches to Assessing Primary Cilia

The use of artificial intelligence (Ai) to analyze images is emerging as a powerful, less biased, and rapid approach compared with commonly used methods. Here we trained Ai to recognize a cellular organelle, primary cilia, and analyze properties such as length and staining intensity in a rigorous and reproducible manner.

Transcript

Current methods of cilia analysis are labor intensive and prone to error and bias. Our approach seeks to streamline time and effort while mitigating potential errors. The main advantage that this technique offers is increased rigor and reproducibility and quantitative image analysis.

This type of approach is not only relevant to cilia analysis, but can be broadly applied to many cell biological questions, including those dealing with other organelles and cytoskeletal proteins. To begin, open the training dataset, select file from the menu, click import export and select create ND file from file sequence. Select the folder containing the training data set, and the list of files will open in the center of the dialog window.

Manually define the organization of files using at least one option in the dropdown menu. Enter the corresponding numerical values under each selected option and select none wherever options are not selected. Click convert to open the ND document.

To calibrate the image, right click on the uncalibrated option in the lower left-hand corner of the image. Click calibrate document, then click pixel size, enter the value and click okay. Select view analysis controls, open the binary toolbar and select auto detect or draw object to hand identify cilia by precisely tracing individual ciliary structures on all opened frames.

For training the AI, select nis. ai, click train segment. ai to open the train segment.

ai box and then select the source channel to be used for training. Select the appropriate GroundTruth binaries to train the AI.Select the required number of iterations to train the AI depending on the binary size and distribution, then select the destination folder to save the trained AI file and click train to train the software. This process takes several hours.

Open the experimental confocal images of cilia as described before by converting the sample. TIF files to ND2 files. In the pop-up window, select multi-point from the first dropdown menu and enter a value corresponding to the total number of the images.

In the second dropdown box, select wavelength and change the value to the total number of the channels in the folder. The software will automatically unlock a wavelength selection window, located at the bottom, right end of the pop-up window. In the wavelength selection window, use the color dropdown menu to select the color of each channel.

Provide each channel with a different name under the name column. Once all information is updated, click convert. Calibrate the images as described before.

Make sure the pixel size of the experimental dataset is consistent with that of the training dataset. Identify cilia on the first channel by using the trained AI from the previous step. Open nis.

ai from the menu, select segment. ai, then select AC3 in the source channels. Then identify the cilia on the second channel by selecting MCHR1 in the source channels.

The software will draw binaries on the labeled cilia. Next check the images for any misidentified binaries. Select delete object in the binary toolbar to manually delete the misidentified binaries.

Once the cilia have been identified and segmented, analyze different cilia parameters, such as lengths and intensities, using the general analysis three tool. Select image from the menu and click new GA3 recipe. A new window with a blank space in the center will be opened.

GA3 will automatically detect the binaries appropriately labeled according to the AI and include the corresponding node. GA3 will also automatically detect the channels in the images and display their tabs under channels. The AI will segment all cilia like objects in the frame and detect incomplete cilia along the edges of the frame.

To remove them, select binary processing, remove objects, and then drag the touching borders node into the blank space and connect the node to the appropriate binaries. To measure cilia length, select measurement object size, and then length. Drag and drop the parameter to the center and connect to the appropriate binary node.

To measure cilia intensities, select some object intensity. Drag and drop the parameter to the center and connect to the appropriate binary node and channel of interest. In the measurement menu, go to object ratiometry and select Mander's coefficient to set up the colocalization pathway in GA3 by measuring the overlap of two channels within individual cilia.

Drag and drop the Mander's coefficient node in the blank space and connect it to the appropriate binary and channels. ApEn the measurements and the single table by opening the data management menu. In the basic category, select ApEn column and then click run now to measure cilia.

All the measurements will appear in a single output table. The representative images show that the trained AI properly identified cilia in vitro in the images of IMCD cells, primary hypothalamic cultures and hippocampal cultures, but no other non ciliary structures such as cytokinetic bridges. The length of the cilia ranged from 0.5 to 4.5 micrometers in IMCD cells and two to 12 micrometers in the hypothalamic and hippocampal culture.

The AI measured the lengths of AC3 labeled cilia in vivo in the images of our arcuate nucleus, paraventricular nucleus and cornu ammonis one regions. According to the analysis, hypothalamic cilia in vivo ranged from one to 15 micrometers, as seen in white and brown bars. While cilia and the cornu ammonis region ranged from one to 10 micrometers, as seen in gray bars.

Interestingly, the intensity of the ciliary MCHR1 was stronger in the paraventricular nucleus than that in the arcuate nucleus. The intensities of MCHR1 against AC3 were plotted to measure their overlap. The majority of cilia were positive for both markers while some cilia were positive for either AC3 or MCHR1.

To quantify colocalization of MTHR1 within AC3, Mander's overlap coefficient was measured and there was a significant increase in the overlap in the paraventricular nucleus than in the arcuate nucleus. To measure intensity along the length of the cilia, cilia polarity was defined using Centrin2-GFP as the basal body marker. This allowed to distinguish the base of cilia from the tips of ARL13B-M cherry positive cilia.

Changes in ARL13B intensity along the length of cilia were observed where ARL13B intensity was higher at the base than at the tip of the cilium in arcuate nucleus, seen on the left, as well as PVN, as seen on the right. When analyzing data with this approach, it is important to make sure that the quality and resolution of the experimental dataset is consistent with that used to train AI.The main utility of this approach is after detecting cilia by AI, the user can be creative about which properties are analyzed by customizing the analysis workflow integrated within the software.

Read Article

Journal (Biology)

Journal (Biology)

Artificial Intelligence Approaches to Assessing Primary Cilia

Chapters

Transcript

Related Videos

Get cutting-edge science videos from JoVE sent straight to your inbox every month.

Your free access has ended.