“Such technology not only has the potential to improve patient diagnosis, but also can be utilised for targeted evaluation of therapy efficacy.”Dr Heba Sailem, Senior Lecturer of Biomedical AI and Data Science
20 November 2023
AI-powered tool promises to help clinicians diagnose cancer
The new tool is able to identify the distinct cellular patterns that characterise mesothelioma subtypes, an aggressive cancer that typically develops due to exposure to asbestos.
A new study, published in Cell Reports Medicine, details how a new AI tool called a ‘MesoGraph’ can help clinicians diagnose mesothelioma.
Mesothelioma is an aggressive type of cancer, with survival rates of less than 10% over a five-year period. It develops in the thin layer of tissue that separates the lungs from your ribcage and chest wall (known as the pleural lining) and it is primarily associated with asbestos exposure.
One way that mesothelioma can be diagnosed is based on histopathological images. A histopathological image is a picture of a very thin slice of tissue that is transparent and can be stained with different colours, enabling researchers to study structures or cells under a microscope. In the case of mesothelioma, scientists focus on changes in cellular morphology and shape. However, diagnosis of mesothelioma remains a major issue, with one challenge being the variation of the phenotype of cancer cells within tumours that can lead to divergent diagnoses among healthcare professionals.
But a new study, involving Dr Heba Sailem and led by the PRISM team, has demonstrated the potential of an AI technology in identifying the distinct cellular patterns that characterise different mesothelioma subtypes with minimal human input. The AI tool employ graph neural networks that consider tumours as a network of interacting cells, capturing the surrounding environment alongside the characteristics of individual cells.
The authors were also able to test its reliability with morphological analyses (an analysis of the cell’s structure and form) of the predicted cell scores, which confirmed its agreement with known differences about the various subtypes of mesothelioma.
An important feature of MesoGraph is that the results of an analysis can be visualised using a user-friendly interactive graphical user interface (GUI). These results highlight the potential of supporting clinicians with analysing such images, toward improving diagnoses for mesothelioma.
This could also lead to more effective treatments. As mesothelioma can develop into one of three different types that each might benefit from different treatments, this analysis can help identify the type of mesothelioma and the best course of treatment.