Researchers Accelerate Teravoxel Image Segmentation and Analysis Using AI Techniques with MATLAB
CUHK Streamlines Workflow, Enabling Comprehensive Cell Typing and Molecular Profiling in a Single Environment
“We found the new development in
blockedImageandcellposein MATLAB to be very timely for our workflow, as we wanted to handle both image processing and segmentation in a single script, alongside other classical image processing algorithms.”
Key Outcomes
- MATLAB with
cellposeandblockedImagecan handle both image processing and segmentation in a single script - Large images are parsed into much smaller stacks with
blockedImage, eliminating the need for expensive, high-end computers and reducing programming time and errors - With Cellpose AI techniques, cell segmentation that once seemed daunting is now possible to accelerate segmentation and analysis of teravoxel images
The ~800 GB whole mouse brain data set with Cellpose segmentation and a scatter plot of each cell’s centroid in MATLAB.
The Chinese University of Hong Kong (CUHK) undertakes a wide range of research programs in many subject areas, including a program to develop methods for efficient probing and mapping biological structures and molecular compositions.
Integral to the method development is image processing that allows flexible handling of multidimensional images and large volumetric data sets. CUHK uses functions in MATLAB®, Medical Imaging Toolbox™, and Image Processing Toolbox™ extensively for this purpose, especially cellpose and blockedImage, accelerating the cell segmentation that the team thought would be daunting and even impossible.
In its recent study, CUHK has two 3D image data sets that are analyzed using cellpose in MATLAB. One has 10 teravoxels and 28 channels representing about 1 million cells that require segmentation and cell typing analysis. The second is an approximately 800 GB whole mouse brain image requiring global neuronal soma segmentation and registration to Allen Brain Atlas.
Dr. Lai’s team at CUHK utilizes cellpose in MATLAB on thresholded, background-subtracted images for segmentation aided by blockedImage to obtain the cell masks before proceeding to analyze the molecular expression profiles of each cell. The obtained cell masks for the 3D 28-plex images allow CUHK to profile the immunostaining intensities of 25 chosen markers, which are used for cell typing classification, all done using MATLAB in a single script.
Researchers at CUHK chose MATLAB because they wanted the whole pipeline in a single environment, and they found the streamlined workflow, good documentation, and good technical support from MathWorks enabled them to handle their large data sets. CUHK hopes to scale the technology further for clinical use, where on-the-fly image processing can lead to more efficient patient diagnostics.
Products Used
Related Resources
