Children's Hospital of Philadelphia Researchers Announce New AI Model for Cell Segmentation and Classification
The PennZone/10288893

Trending...
PHILADELPHIA, Nov. 25, 2024 ~ A team of researchers at Children's Hospital of Philadelphia (CHOP) has developed a new AI technology called CelloType, which aims to improve the accuracy of identifying and classifying cells in high-content tissue images. The findings of this study were published today in the journal Nature Methods.

Spatial omics is a rapidly growing field that combines molecular profiling with spatial information to map where different molecules are located within cells in complex tissues. This approach provides detailed insights into how diseases develop and progress at the cellular level, aiding in the development of precise diagnostics and targeted treatments. CHOP has been at the forefront of translational research in this area, using spatial omics to study a wide range of complex diseases such as cancer and chronic kidney disease.

The first step in analyzing spatial omics data is cell segmentation, which involves identifying cell boundaries, followed by cell classification. Recent advancements in technology have allowed for the analysis of intact tissues at the cellular level, providing unprecedented insights into the link between cellular architecture and tissue functionality. As a result, CHOP is currently involved in several high-profile projects such as the Human Tumor Atlas Network, Human BioMolecular Atlas Program (HuBMAP), and BRAIN initiative, all of which use similar technologies to map spatial organizations of healthy and diseased tissues.

More on The PennZone
Lead author of the study and professor in the Department of Pediatrics at CHOP, Dr. Kai Tan stated that they are just beginning to unlock the potential of this technology. He believes that this approach could revolutionize our understanding of complex tissues at the cellular level and pave the way for transformative breakthroughs in healthcare.

With an increasing amount of spatial omics data being generated, there is a pressing need for more sophisticated computational tools for data analysis. This led Dr. Tan and his team to develop CelloType, which utilizes transformer-based deep learning - a type of AI that can capture complex relationships and context within high-dimensional data. This makes it highly efficient for handling large-scale tasks such as natural language processing and image analysis, allowing it to improve accuracy in cell detection, segmentation, and classification.

In this study, the researchers compared CelloType with traditional methods using animal and human tissue datasets. They found that CelloType's multi-task learning strategy, which integrates segmentation and classification simultaneously, was more efficient than the typical two-stage approach. Additionally, CelloType outperformed existing segmentation methods on various types of images.

More on The PennZone
When it came to cell type classification, CelloType surpassed a model comprised of state-of-the-art methods for individual tasks and a high-performance instance segmentation model. The latter uses AI to precisely outline objects in an image. The researchers also demonstrated how CelloType can be used for multi-scale segmentation and classification of both cellular and non-cellular elements in a tissue using a multiplexed tissue image - an advanced biomedical image that displays multiple biomarkers within a single tissue sample. This allowed for detailed analysis of both small and large cell structures.

Dr. Tan emphasized the pivotal role technology plays in today's biomedical research and stated that CelloType advances spatial omics by providing a robust and scalable tool for analyzing complex tissue architectures. This will expedite discoveries in cellular interactions, tissue function, and disease mechanisms.

Researchers outside of CHOP can access CelloType through open-source software in a public repository for non-commercial use. This research was supported by grants from the National Cancer Institute (NCI) Human Tumor Atlas Network (#U2C CA233285) and the National Institutes of Health (NIH) Human Biomolecular Atlas Program (#U54 HL165442).

The study titled "CelloType: A Unified Model for Segmentation and Classification of Tissue Images" was published online on November 22, 2024, in Nature Methods with DOI: 10.1038/s41592-024-02513-1.
Filed Under: Business
Show All News | Report Violation

0 Comments

Latest on The PennZone