Contact Us

AI-Powered Multiplexed Image Analysis to Explore Colon Adenocarcinoma

Understanding cancer progression using an AI-powered multiplexed image analysis-based exploration of the tumor immune microenvironment in colon adenocarcinoma

 Human_colon_adenocarcinoma_CAC_tissue.jpg

Colorectal cancer is the second leading cause of cancer deaths worldwide, and studying the colon cancer immune microenvironment is necessary to improve treatment outcomes. In this study, we demonstrate a spatial biology workflow utilizing Cell DIVE and AIVIA software to map the tumor immune landscape in colon adenocarcinoma. By visualizing 30 biomarkers within a single tissue and performing AI-powered segmentation and phenotyping, we are able to uncover multiple facets of tumor cell aggression.

Key Learnings

  • Discover how whole-tissue multiplexed imaging using Cell DIVE can map the tumor-immune microenvironment.
  • Explore the complex interplay of tumor, immune and stromal cells involved in tumor progression.
  • Learn how AI-powered analysis workflow using Aivia software can be utilized to unveil hidden spatial insights into tumor cell aggression.

Accurately detect cells in multiplexed images using AI

To draw quantitative, biological insights the first step is often detecting and segmenting the individual cells within the entire image. Aivia’s Multiplexed Cell Detection recipe leverages a modified Cellpose [1] algorithm to accurately detect cell nucleus and membrane with different morphologies. In this CAC example, the DAPI and NaKATPase channels were used as a nucleus and membrane marker respectively to segment all the cells in this tissue. Within the CAC tissue, a total of 555,480 cells were identified and segmented, with an average cytoplasmic area of 66.51 µm2, average nuclear area of 38 µm2, and nuclear circularity of 0.77.

Automatic clustering to reveal unknown phenotypes

Unknown cell phenotypes can also be identified in an unsupervised way using the automatic clustering tool in Aivia. Instead of defining a priori biomarker profile for each phenotype, the PhenoGraph-Leiden [2] clustering tool can create groups of phenotypes based solely on biomarker expression similarity across the entire panel. These phenotypic clusters can then be organized based on their similarity to the other clusters discovered in the analysis via heat map and dendrogram. Utilizing the dendrogram feature we can visually examine some important trends and enriched cell populations within the typically heterogenous CAC dataset. Additionally, cell phenotypes that are relatively less abundant in the tissue may be captured with ease with such analyses.

Conclusion

The complexity of spatial biology datasets brings unique challenges to data collection, annotation, and analysis. The process detailed here: segmentation, phenotyping, and clustering of phenotypes is one pathway into extracting useful information from large 2D datasets such as Cell DIVE images. By following this general pathway, researchers can extract cell identity, cell position, morphological data, and more while applying that information directly to their tissue of interest. Beyond raw analysis, clustering and dimensionality reduction-based approaches allow for both a high-level overview of the data and the possibility of hypothesis generation based on unforeseen relationships in phenotypic clusters.

Download the Application Note

By clicking on the SUBMIT button, I confirm that I have reviewed and agree with Leica Microsystems GmbH Terms of Use and their Privacy Policy. I also understand my privacy choices as they pertain to my personal data, as detailed in the aforementioned Privacy Policy under ‘’Your Privacy Choices’’.

Additional links of interest

Scroll to top