Breast cancers can now be better diagnosed after researchers from Massachusetts Institute of Technology (MIT), ETH Zurich, and the University of Palermo in Italy developed an artificial intelligence (AI) model that can determine breast tumor stages—a breakthrough that can save patients from overtreatment.

The latest model, now published as open-access research in Nature Communications, revealed that both the state and arrangement of cells in a breast tissue sample play a crucial role in identifying the stage of ductal carcinoma in situ (DCIS).

DCIS is a preinvasive tumor that accounts for around one-quarter of all breast cancer diagnoses. Since the biomarkers that specify whether or not it will progress into invasive cancer are still unknown, women suffering from this tumor are often overtreated.

With the development of the AI model, clinicians can diagnose simpler cases of DCIS faster and cheaper, allotting more time for complicated ones to be evaluated and protecting patients from undergoing excessive treatments.

Cell States as Breast Cancer Indicator

Most of the existing diagnostic tests in determining the stage of DCIS are expensive. These include single-cell RNA sequencing and multiplexed staining.

MIT researchers Caroline Uhler, lead author Xinyi Zhang, ETH Zurich professor GV Shivashankar, and the rest of their team sought to create a more affordable technique that could be as detailed as pricier methods.

They accomplished that by fusing a cheaper imaging approach, called chromatin staining, with an AI model.

Using a large dataset containing 560 tissue sample images from 122 patients at three different stages of breast cancer, the researchers trained the AI model to learn the representation of the state of each cell in a tissue sample image. This knowledge is then used by the AI to identify the stage of breast cancer.

However, since not all cells can point to cancer, they also had to build the model to aggregate cells into clusters with similar states. Eight states were discovered to be significant markers of DCIS.

Some of these states are more indicative of invasive cancer. Thus, the AI model determines the proportion of cells in each state for a single breast tissue sample.

Cell Arrangement for Better Accuracy

In addition to proportion, the researchers designed the AI model to account for the arrangement of cell states, improving the accuracy of the diagnosis.

According to co-corresponding author Shivashankar, relying on the proportions of cells in every state is insufficient in diagnosing the stage of DCIS since the organization of cells also varies.

“The interesting thing for us was seeing how much spatial organization matters. Previous studies have shown that cells which are close to the breast duct are important. But it is also important to consider which cells are close to which other cells,” added Zhang.

To confirm the model’s accuracy, they compared its results to the samples evaluated by a professional pathologist. The comparison showed positive outcomes after clear agreements were revealed in multiple cases.

Otherwise, the model could offer details about the tissue sample, such as the arrangement of cells, which human experts could consider in making a diagnosis.

This capability allows the AI model to be beneficial not just for breast cancer patients but also for those who are suffering from other types of cancer. The researchers are also looking at the potential to use this technology in neurodegenerative conditions.