Researchers from the Massachusetts Institute of Technology (MIT) and ETH Zurich have developed an artificial intelligence (AI) model that can identify certain breast cancer stages and determine whether it is likely to progress into invasive cancer.
Published in Nature Communications, the new model could help clinicians assess the type and stage of ductal carcinoma in situ (DCIS) and reduce overtreatment.
Accounting for around 25% of all breast cancer diagnoses, DCIS is a difficult-to-treat, preinvasive tumour that progresses to a highly invasive stage of cancer in up to 50% of patients.
Currently, researchers use techniques such as multiplexed staining or single-cell RNA sequences to determine the stage of DCIS in tissue samples. However, these tests can be expensive to perform.
Using 560 easy-to-obtain breast tissue sample images from 122 patients at three different stages of disease that were taken using an imaging technique called chromatin staining, researchers trained the AI model to learn the representation of the state of each cell in the sample image to determine the stage of a patient’s cancer.
As some cell states were more indicative of invasive cancer than others, researchers aggregated these cells by designing a model to create clusters of cells in similar states, identifying eight states that were recognised as important markers of DCIS, to determine the proportion of cells in each state in a single tissue sample, which significantly boosted its accuracy.
The model had clear agreement in many instances compared to samples evaluated by a pathologist and could provide information about tissue sample features such as the organisation of cells.
In addition, the model could be adapted for use in other types of cancer and even potentially neurodegenerative conditions.
Caroline Uhler, researcher at MIT’s Laboratory for Information and Decision Systems, commented: “We took the first step in understanding that we should be looking at the spatial organisation of cells when diagnosing DCIS and now we have developed a technique that is scalable.
“There is still much more research to do, but we need to take the organisation of cells into account in more of our studies.”