A revolutionary artificial intelligence tool that acts like “sonar” for cancer could pave the way for highly personalised patient care by analysing complex tissue samples in just one minute.
Researchers at the University of Cambridge have developed a machine learning algorithm capable of identifying distinct tumour subtypes and levels of aggressiveness at unprecedented speed, offering a solution to the time-consuming bottleneck of manual pathology.
The study, published in Nature Cancer, demonstrates that the tool — named SMMILe — can predict tumour behaviour and location without requiring detailed, labour-intensive annotations during training.
“Cancer isn’t always uniform. A single tumour can contain different subtypes, some that are more aggressive than others,” said Dr Zeyu Gao from the Early Cancer Institute. “Our model doesn’t just say ‘yes, there’s cancer’, it maps out these subtypes and their proportions within the tissue. This could one day help doctors tailor treatments more effectively, moving to a more nuanced understanding of each patient’s cancer.”
Seeing in the dark
The algorithm was tested on 3,850 whole-slide images across six cancer types, including lung, breast and prostate cancer. By learning from slides marked only with simple diagnostic labels, SMMILe matched or exceeded the performance of nine state-of-the-art tools that require far more detailed training data.
“What we’ve developed is akin to a ‘sonar’ for images that essentially allows us to see in the dark,” said joint senior author Dr Mireia Crispin-Ortuzar. “With our new AI method, we can accurately map the tumour samples – and the beauty is that it is trained on cheap, widely-available datasets that only contain bulk, non-spatial information”.
The team plans to expand the tool to predict biological signatures that reveal how tumours behave at a molecular level, potentially accelerating the delivery of targeted therapies.
