Using Computer Vision Libraries to Streamline Nuclei Quantification

We developed this method to normalize metabolic data from cell models to help identify mechanisms, underline heat therapy induced skeletal muscle adaptations, and ultimately, improve metabolic health among people with pre-diabetes.

We must count nuclei for experimental normalization. Manual quantification of nuclei presents challenges including observer bias, time, and variability in coming across different samples or conditions.

Our program is open source ensures usability by scientists with various levels of coding-related technological skills, and is validated for the specific task of quickly and accurately quantifying nuclei.

This technique allows us to objectively validate the mechanisms underlying the potential effect of heat therapy on muscle and mitochondrial health benefits from our recent NIA-funded clinical study.

[Narrator] To begin, launch a web browser on a computer system, navigate to github.com and nuclei counter releases. Download the latest version of the file named Count nuclei.zip. From the Downloads folder, right click on the zip file and select Extract all to extract the files to the desired location on the local computer. Next, search for CMD or command prompt in the search bar to open a command prompt. Use the CD command to change the directory to the file path of the executable file, which is the application file that was just extracted from the download folder. Then press Enter to confirm the directory change. On the next command line, replace path to images with the file path to the folder containing images to be analyzed. Path to output with the file path to the folder where the .csv file should be saved and results.csv with the desired file name for the output. An example code is shown on the screen and the file paths to images and output can be inserted as shown in the quotation marks. Use results.csv as the results file name or specify another. Then, press Enter. When the next command line appears, confirm that processing is complete. Verify that the contours and the results spreadsheet are available in the specified output directory. Visually inspect the contours and compare with counts to verify count quality before data normalization. Open a browser and navigate to the nuclei counter on github.com. Click the green Code button, then select Download ZIP to download the code repository. For Mac OS, click on the file menu from the Downloads folder and select Open to extract the files to the local computer. Navigate to the extracted folder named nuclei_counter main, which contains the code repository. Save the folder in an accessible location and note the file path in a text document. Next, press Command + Space bar to open Spotlight. Then type terminal in Spotlight and select the terminal application. Use the CD command to change the directory to the code repository path by copying and pasting the file path from the text document and press Enter. On the next command line, ensure there is a space after the dollar sign. Then type the given command and press Enter to install the required libraries and enable editable mode. Include the appropriate Python version immediately after pip as shown without a space. Type the onscreen command on the next command line to change the directory to the main source code directory, which is the CD nuclei counter as shown on the screen. Then, type the onscreen command replacing file paths as appropriate, and press Enter. When the next command line appears, confirm that processing is complete. Verify that the contours and the results spreadsheet are available in the specified output directory. Visually inspect the contours and compare with counts to verify count quality before data normalization. All nuclei in images generated by the automated program were outlined by solid green contours indicating that the nuclei were successfully counted. Inter-rater reliability between the two manual counts was excellent with an intraclass correlation coefficient greater than 0.999 and P-value less than 0.0001. The automated program demonstrated excellent reliability when compared to the average manual count with an intra-class correlation coefficient of 0.993 and P-value less than 0.0001. Excellent reliability was observed across all cell density quartiles with intra-class correlation coefficients ranging from 0.986 to 0.998 all with P-values less than 0.0001. Areas with multiple nuclei clustered together or areas with an artifact such as a halo were not accurately counted by the automated program. These potential problems along with possible causes and troubleshooting steps to improve both image quality and the accuracy of the automated nuclei quantification workflow are listed in the table on screen.