As highly multiplexed imaging becomes increasingly prevalent in digital pathology, researchers are generating more complex datasets than ever before. Image analysis results containing dozens of markers across millions of cells are now common and have incredible potential for increasing our understanding of tumor microenvironments, immune landscapes, and cell-cell interactions. However, the complexity of these datasets brings new challenges. Extracting meaningful biological insights requires not only powerful image analysis tools but also intuitive methods to navigate high-dimensional data and identify key phenotypic patterns across cell populations. We’re excited that with the launch of our new High Dimensional Analysis module, all of these capabilities can be leveraged natively by HALO® users. 

Throughout the development of our HALO image analysis platform, support for researchers working with highly multiplexed data has been a core focus. The platform offers a robust selection of tools for segmentation, biomarker quantification, and spatial analysis. With the ability to generate and export object-level data, HALO users can also leverage third-party software for external clustering analysis. External results can then be imported back into HALO to overlay cell phenotypes on whole slide images or perform spatial analysis using the HALO Spatial Analysis module. This export-import workflow supports high dimensional workflows but relies on users switching between platforms. 

With the latest version of HALO, the recently released HALO 4.1, users working with highly multiplexed images can leverage powerful new capabilities using the High Dimensional Analysis module. The module offers dimensionality reduction and unsupervised clustering with interactive plotting natively within HALO, helping researchers explore and understand complex datasets faster. In this blog, we’ll cover the basics of dimensionality reduction and unsupervised clustering before exploring the High Dimensional Analysis module in more depth. 

An Introduction to Dimensionality Reduction and Unsupervised Clustering 

In quantitative image analysis for pathology, morphological and expression data is measured on a cell-by-cell basis across entire tissue sections. When each cell in an image is quantified for potentially dozens of biomarkers, the result is a high-dimensional dataset where every cell is described by numerous features. Analyzing these datasets manually can be laborious and time intensive, and important details can be missed in a sea of data. 

Dimensionality reduction techniques help address this challenge by reducing the number of variables under consideration while preserving the relationships between data points. This makes it possible to visualize complex patterns in two dimensions and explore trends that might otherwise go unnoticed. While there are multiple dimensionality reduction techniques, the High Dimensional Analysis module uses Uniform Manifold Approximation and Projection (UMAP). UMAP is a relatively fast technique, scalable, and effective at preserving both the global and local structure of data, making it a powerful tool for revealing cellular phenotypes and population structure in multiplexed imaging data. 

In unsupervised clustering, objects are grouped into clusters based on similarities in their underlying features. This technique allows researchers to discover phenotypes that were not pre-defined—such as rare immune subpopulations or transitional states between cell types. 

The High Dimensional Analysis module offers two powerful clustering methods: 

• K-means, which is fast and efficient but requires the user to define the number of clusters. 

• PhenoGraph, which is more computationally intensive but does not require pre-specifying the number of clusters, making it well suited for discovering subtle population structures. 

Together, dimensionality reduction and unsupervised clustering provide a robust framework for phenotyping cells in complex datasets. 

Capabilities of the High Dimensional Analysis Module