AI-based Multiplex Image Analysis of Pathology Slides

Background

Recent advances in single-cell sequencing are revealing the complexity of heterogeneous cell populations within the tumour microenvironment. The abundance and spatial location of subsets of cells have been linked with tumour behaviour and the response to therapies, highlighting the need to carry out detailed mapping of single cells within the tumour microenvironment to improve understanding of their roles during disease progression.

The development of multiplexed imaging techniques, such as multiplexed immunohistochemistry (mIHC), multiplexed immunofluorescence (mIF) and imaging mass spectrometry (IMC) have enabled researchers to map the spatial distribution of proteins and transcripts within tumour samples.

A common task following multiplexed imaging is to identify specific cell types marked by antibody staining. However, the manual annotation of individual cells by trained pathologists is impractical to carry out at any meaningful scale. While automated tools for cell identification in histopathology images have been developed, there is currently a lack of accurate, generalisable and scalable approaches to analyse pathological sections at the single-cell level.

Technology Overview

Digital pathology is an emerging field. It uses sophisticated computing tools and AI to diagnose disease and guide treatments faster and more easily – offering exciting opportunities for the development of novel ways to understand cancer.

The Computational Pathology and Integrative Genomics Team at the ICR, has created an AI-based approach that provides rapid accurate and reliable analysis of multiplex histopathology slides at a single cell level with minimal human input.

Self-supervised Antigen Detection AI (SANDI) involves scanning slides stained with multiplexed labels into digital images and using an AI algorithm to analyse the images automatically at a single-cell level. Developed and validated using datasets from normal and tumour samples (breast, ovarian cancer and myeloma), the technique is agnostic to the antigen detection method or imaging platform used. In principle, it should detect an unlimited number of antigens within the same cell and work across all types of cancer and cell types.

Benefits

• Universal and quantitative automation of single-cell detection from multiplex pathology slides with minimal human input.
• Agnostic to the antigen detection method and imaging platform used to stain and visualise tissue samples.
• Pioneering use of AI in digital pathology.

Applications

The technology could help accelerate research into how different cell types within the tumour microenvironment influence tumour behaviour, prognosis and response to treatment.

Opportunity

The Institute of Cancer Research, London, is now seeking licensees and commercial development partners looking to apply this technology in R&D programmes for an AI-based method enabling the automated, rapid and accurate identification and classification of single cells from multiplexed pathology slides.