Ductal carcinoma in situ (DCIS) is a preinvasive tumor that accounts for approximately 25% of all breast cancer diagnoses. Determining which DCIS cases will progress to invasive cancer poses a significant challenge for clinicians, often leading to overtreatment. Now, an interdisciplinary team from MIT and ETH Zurich has developed an AI model that could revolutionize DCIS diagnosis. This model analyzes readily available breast tissue images to identify different stages of DCIS, showing that both the condition and arrangement of cells within the tissue are crucial for accurate staging.
The research team utilized a large dataset of tissue images, leveraging their accessibility to train and test their AI model. The model’s predictions were then compared to the conclusions of a pathologist, revealing a strong correlation in many cases.
We took the first step in understanding that we should be looking at the spatial organization of cells when diagnosing DCIS, and now we have developed a technique that is scalable. From here, we really need a prospective study. Working with a hospital and getting this all the way to the clinic will be an important step forward,
explains Caroline Uhler, a professor in MIT’s Department of Electrical Engineering and Computer Science and a co-author of the study.
The potential impact of this AI model is significant. By accurately identifying DCIS stages, clinicians can streamline the diagnosis of less complex cases, reserving labor-intensive tests for situations where the risk of invasive cancer is less clear.
The study highlights the importance of combining imaging with AI. While techniques like multiplexed staining and single-cell RNA sequencing can determine DCIS stages, their high cost limits their widespread use. This new model, however, utilizes a cost-effective imaging technique called chromatin staining, which, when combined with machine learning, can provide comparable information about cancer stage.
The model analyzes tissue sample images, learning to represent the state of each cell and using this information to infer the stage of the patient’s cancer. Recognizing that not all cells are indicative of cancer, the researchers designed the model to cluster cells in similar states, identifying eight key markers of DCIS. The model then determines the proportion of cells in each state within a tissue sample.
Interestingly, the research revealed that the organization of cells is as important as their individual states. “But in cancer, the organization of cells also changes. We found that just having the proportions of cells in every state is not enough. You also need to understand how the cells are organized,” says GV Shivashankar, a professor of mechano-genomics at ETH Zurich and co-author of the study.
By considering both the proportion and arrangement of cell states, the model’s accuracy was significantly improved. This spatial awareness allows the model to provide valuable insights into tissue sample features, such as cell organization, which can aid pathologists in their decision-making process.
The versatility of this AI model extends beyond DCIS, with potential applications in other types of cancer and even neurodegenerative conditions. As Uhler emphasizes, “We have shown that, with the right AI techniques, this simple stain can be very powerful. There is still much more research to do, but we need to take the organization of cells into account in more of our studies.”
This groundbreaking research, funded by organizations including the Eric and Wendy Schmidt Center at the Broad Institute, ETH Zurich, and the U.S. National Institutes of Health, was published in Nature Communications on July 20th. It represents a significant step towards improving DCIS diagnosis and treatment, potentially leading to better outcomes for patients.
Responses (0 )