With over 100 subtypes, diagnosing sarcoma correctly can be very challenging. Sarcoma pathologists analyse samples of sarcomas to look for features which point to different subtypes – but this can take a long time. Making a correct diagnosis is vital as treatment and prognosis can differ between different subtypes.
We need ways of diagnosing sarcoma quicker and more effectively – and that’s where artificial intelligence (AI) can play a role. Information that the pathologist does not recognise as important can be identified by AI and predict response to treatment. This means it can reduce diagnostic errors, speed up diagnosis and use less tissue that could be made available for other important tests – meaning more benefit for patients.
For AI models to be successful, it is important to have images that have been correctly labelled by expert pathologists. The information becomes more powerful when linked with patients’ clinical and genomic data. Computer scientists can use this information for the development of better AI models. AI is already being effectively in more common cancers, such as breast cancer and melanoma, but little has been done on sarcoma – which is what the team aim to change.
How will this project tackle this challenge?
As a first step, the team will scan slides containing soft tissue sarcoma samples from about 4000 patients diagnosed since 2000, which will help improve the data available for future studies.
Computer scientists and pathologists will train computer models to tell the difference between six types of soft tissue tumours. They will then evaluate if their computer model is as good or even better than pathologists.
To ensure that the results gathered from this work will benefit as many people as possible, this project will bring together about 30 pathologists from across the UK and beyond, to improve diagnostic clinical service for patients by working on AI. The scanned slides can also be used for training future pathologists.
What this means for people affected by sarcoma
The team hope that by developing an AI model to improve diagnosis, fewer special tests will be required of patients. This should make diagnosis quicker, more efficient, and ensure that more resource is available to ensure patients get the right sarcoma diagnosis.
With over 100 subtypes, diagnosing sarcoma correctly can be very challenging. But making a correct diagnosis is vital as treatment and prognosis can differ between different subtypes.