ORIGINALLY PUBLISHED:
4 May 2023
Written by:
Executive Director, Computational Pathology, Cancer Biomarker Development
Vice President, Cancer Biomarker Development
Vice President, Image Data Sciences, Computational Pathology, Oncology R&D
At AstraZeneca, we’re pioneering new computational pathology approaches, combining digital pathology and big data with cutting-edge artificial intelligence to enhance patient selection and enable more personalised treatments, with the ultimate goal of improving patient outcomes.
Beyond the microscope: computational pathology
In 1838, German pathologist Johannes Müller published the first known description of cancers observed down a microscope.1 Nearly two centuries later, the basic process for cancer diagnosis remains more or less the same - slices of tumour tissue fixed, stained and magnified for human viewing - although the tools and technologies have improved significantly since Müller’s time.
Over the past two decades we have seen the advent of technologies, such as high-resolution slide scanning, creating digital images of immunohistochemically stained slides that can be easily viewed and shared.2
Advances in data science and computing now allow us to integrate digital pathology images with genomic, radiological, clinical and other data, applying deep learning algorithms to gain novel insights.
By extracting as well as analysing objective and clinically relevant information from these complex data sets, we can ‘see’ more than the human eye ever could to understand disease and guide treatment.
Known as computational pathology, this approach uses the power of artificial intelligence to enhance patient selection in oncology3 – a key part of our precision medicine strategy.
Discover how we are utilising computational pathology to revolutionise the development of cancer therapies and diagnostics in the video below:
Transformational medicines need transformational diagnostics
As our understanding of disease and the options to treat it increases, so does the need for a precision medicine approach to ensure that we target patients with the right medicine, earlier in their disease.
For example, antibody-drug conjugates (ADCs) rely on the unique targeting ability of monoclonal antibodies to kill cancer cells and reduce damage to normal cells.4 Similarly, immunotherapies are based on knowledge about the tumour microenvironment and the defensive mechanisms that cancer cells use to escape the immune system.5
Conventional pathology depends on manual, subjective scoring of tissue biomarkers to help us understand disease progression and select the best treatments for patients. In addition, it is usually based on a limited subset of cells within a sample.
Computational methods that capture more quantitative information about underlying tumour biology are critical for supporting the evolving requirements of clinical decision-making when it comes to these novel therapies.6
The computational pathology tools we are developing at AstraZeneca allow us to analyse hundreds of thousands of cells per slide in unprecedented detail, much faster than previously possible.
This generation of rich datasets gives us an enhanced view of what’s happening inside a tumour, quantifying tissue-based biomarkers to help us select patients most likely to respond to treatment.
In addition to bringing benefits to patients by providing more in-depth and accurate diagnoses, this technology will also free up pathologists’ time to focus on more complex cases.
Quantitative Continuous Scoring: a new frontier in diagnostics
Quantitative Continuous Scoring, or QCS, is our novel, fully supervised computational pathology solution.3,7
As applied to digitised images of patient tissue samples, QCS can precisely and efficiently quantify targets both on the surface and inside a tumour cell. It is hypothesised that this level of intracellular detail could be used to predict patient responses to therapies like ADCs which are ultimately internalised by tumour cells.8
This hypothesis was tested in a 2024 retrospective and exploratory analysis of the results of a Phase III trial. The analysis showed that the expression of a target of interest, as measured by QCS, was predictive of clinical outcomes in patients with advanced non-small cell lung cancer (NSCLC).9
Historically, when assessed using conventional methods, the same target had not been predictive of patient responses to treatment. The analysis suggested that this more precise quantitative measurement of a target, as enabled by QCS, could substantially improve patient selection for corresponding targeted treatments.
Additional research is required – and ongoing – to further understand the potential utility of QCS in clinical practice and other cancer types such as liver cancer.
Seeing more from every slide
Computational pathology enables us to go much further than a single snapshot of a slide, moving towards a multi-layered, fully integrated ‘geographical map’ of the tumour and its microenvironment. Our ambition is to apply these tools across our portfolio, supporting drug development and enhancing patient selection for clinical trials in oncology.
It’s an incredibly exciting time for the field. We have new tools, novel drugs and digital approaches that are opening up access to this technology. We’re also seeing growing enthusiasm and positive moves in the regulatory environment for these novel diagnostics, suggesting the time is right for a step change.
Collaboration with pathologists, instrument manufacturers and the wider pathology community is critical to establishing routine, AI-driven diagnostics and making this powerful technology more widely available for the benefit of patients in clinical trials. We are working with global partners to help establish the diagnostic laboratory infrastructure needed to make this happen.
At AstraZeneca, we believe that computational pathology will underpin the development of the next generation of cancer therapies and diagnostics, changing the way we do clinical research leading towards improved patient outcomes.
