AI in Tissue Cytometry

Alex Barang & Felicitas Mungenast explore tissue cytometry – also known as next-gen digital pathology

Tissue cytometry can be defined as the in-situ identification and quantification of molecular marker expression, cellular phenotypes, mRNA, multicellular tissue entities, etc. within the native tissue environment. Tissue cytometry is equivalent to flow cytometry in terms of phenotypical and functional analysis, with the advantage of retaining tissue integrity.

Typically, the first step in a tissue cytometry workflow requires the automated digitisation of immunofluorescence or immunohistochemistry processed tissue slides by a scanning system. The second and arguably the most critical step is to perform quantitative computer-assisted image analysis on the digitised slides. The types of research questions that tissue cytometry can address cover a wide range of applications, including molecular single-cell profiling (FISH, RNA-ISH, etc.), quantification of cellular pathogens (viruses, bacteria, parasites), immunophenotyping for determining the immune status in-situ, and the characterisation of cellular sub-populations in spatial context. The goal of tissue cytometry is to attain accurate observer-independent, reproducible and standardised results in an automated fashion in research and clinics.

In the Scope of Precision Medicine

The state-of-the-art in biomedical research and medical/analytical technologies, including tissue cytometry, directly contributes to the increasing efficacy of precision medicine over time. Precision medicine, in particular precision cancer diagnostics, uses applications such as tissue cytometry to determine the molecular processes that trigger diseases or lead to the progression of a disease in a specific patient, thereby providing valuable information for diagnosis, prognosis and adequate treatment strategies. Nowadays, for an individual patient, entire panels of biomarkers/cellular phenotypes as well as their spatial relationships must be assessed, whereas in classical histopathology a patient’s diagnosis depends on the visual examination of an H&E stained tissue slide. On one hand, our ongoing discoveries of novel biomarkers lead us to a potentially higher diagnostic precision and more optimised therapies for patients, and on the other hand, we have to face the challenge of increasing our analytical and diagnostic capacity. The way forward is to deliver innovations in the form of automated quantification and interpretation through computer-assisted analysis and decision support systems.

How Can AI Enhance Tissue Cytometry?

Artificial intelligence (AI) in the context of tissue cytometry can be characterised as machine learning/deep learning models, i.e. algorithms specialising in pattern recognition applied to microscopic images of histological samples. Over the past decades these tools have continuously evolved to become more robust, requiring minimal user input for the automated detection and classification of complex tissue structures/entities. AI tools such as machine learning-based tissue classifiers (i.e. automated recognition of histological structures) and deep learning neural network-based nuclei/single cell segmentation, can be leveraged to advance the speed and accuracy of tissue cytometry. Combining these AI tools can make what was once considered an analysis too complex for the human mind attainable.

The ability to rely on an autonomous and robust deep learning-based algorithm for nuclei segmentation especially suited for challenging tissue environments (high cellular density, weak staining intensity) can be seen as the basis of a successful tissue analysis. An accurate nuclei segmentation is used as a starting point for building more sophisticated tissue analyses (high-content phenotyping, spatial dynamics, etc.).

For higher degrees of automation, machine learning-based tissue classifiers focus on the segmentation of morphological substructures within a tissue section. A model is created by marking just a few areas representative of the specific morphological entities of interest (tumour area, blood vessels, immune cell clusters, etc.) directly on the digitised slide. This model will be specialised for separating the tissue into specific classes and will automatically generate a binary mask for further measurements. To further analyse spatial interdependencies, identified cellular phenotypes can be located in reference to detected tissue classes and other cellular phenotypes.

How can tissue cytometry enrich your research?

TissueGnostics combines a high level of automation and expertise in both imaging and analysis and is thereby able to provide a holistic tissue cytometry solution that covers everything from whole-slide acquisition to high-content image analysis and multidimensional data mining. Imaging modalities offered by the company are multi-faceted, including high-throughput whole-slide scanning in brightfield, fluorescence, multispectral, multiplexing and confocal modes. 

Taking advantage of the recent advancements being made in AI, a machine learning-based tissue classifier and a deep learning-based nuclei segmentation algorithm are integrated into TissueGnostics’ contextual tissue analysis software, StrataQuest, with the goal of making sophisticated tissue analysis more accessible. As AI continues to progress, the company is actively involved in shaping the future of digital medicine by participating in multinational research projects, focusing on the development of new machine learning algorithms (such as the HELICAL project). Yet, the company not only develops novel algorithms based on existing AI technologies; it also actively contributes to the development of novel AI technologies that will constitute disruptive, GDPR- and IVD-compliant intelligent products/tools for future precision medicine, such as the s3ai project.

Alex Barang & Felicitas Mungenast are with TissueGnostics

Recent Issues