Ductal carcinoma in situ (DCIS) is a preinvasive tumor type that sometimes develops right into a fatal type of breast cancer. It accounts for about 25 percent of all breast cancer diagnoses.
Because it’s difficult for doctors to find out the kind and stage of DCIS, patients with DCIS are sometimes overtreated. To solve this problem, an interdisciplinary team of researchers from MIT and ETH Zurich developed an AI model that may discover the several stages of DCIS using a reasonable and simply available breast tissue image. Their model shows that each the condition and arrangement of cells in a tissue sample are necessary in determining the stage of DCIS.
Because such tissue images are really easy to acquire, the researchers were in a position to create one among the most important datasets of its kind, which they used to coach and test their model. When they compared the predictions with a pathologist's conclusions, they found clear matches in lots of cases.
In the long run, the model could possibly be used as a tool to assist clinicians diagnose simpler cases without the necessity for labor-intensive testing, leaving them more time to guage cases where it’s less clear whether DCIS will develop into invasive.
“We've taken step one in understanding that we should always have a look at the spatial arrangement of cells when diagnosing DCIS, and now we've developed a scalable technique. From here, we actually need a prospective study. Collaborating with a hospital and moving this study into the clinic will probably be a vital step forward,” says Caroline Uhler, a professor within the Department of Electrical Engineering and Computer Science (EECS) and the Institute for Data, Systems, and Society (IDSS), who can also be director of the Eric and Wendy Schmidt Center on the Broad Institute of MIT and Harvard and a researcher at MIT's Laboratory for Information and Decision Systems (LIDS).
Uhler, co-author of a paper on this research, is supported by lead creator Xinyi Zhang, a PhD student at EECS and the Eric and Wendy Schmidt Center; co-author GV Shivashankar, Professor of Mechogenomics at ETH Zurich jointly with the Paul Scherrer Institute; and others at MIT, ETH Zurich, and the University of Palermo in Italy. The open access research was published on 20 July in .
Combining imaging with AI
Thirty to 50 percent of patients with DCIS develop highly invasive stages of cancer, but researchers lack biomarkers that would tell a clinician which tumors will progress.
Researchers can use techniques similar to multiplex staining or single-cell RNA sequencing to find out the stage of DCIS in tissue samples, but these tests are too expensive to perform on a big scale, Shivashankar explains.
In previous work, these researchers showed that a low-cost imaging technique called chromatin staining might be as informative because the far more expensive single-cell RNA sequencing.
For this research, they hypothesized that combining this single stain with a fastidiously designed machine learning model could provide the identical details about cancer stage as more costly techniques.
First, they created a dataset of 560 tissue sample images from 122 patients at three different stages of disease. Using this dataset, they trained an AI model that learns a representation of the state of every cell in a tissue sample image to infer a patient's cancer stage.
However, since not every cell indicates cancer, the researchers needed to aggregate them in a meaningful way.
They developed the model to create clusters of cells in similar states and identified eight states which can be necessary markers for DCIS. Some cell states are more indicative of invasive cancer than others. The model determines the proportion of cells in each state in a tissue sample.
Organization is significant
“However, in cancer, the organization of cells also changes. We found that it shouldn’t be enough to simply know the proportions of cells in each state. You even have to grasp how the cells are organized,” says Shivashankar.
With this information, they designed the model to bear in mind the proportions and arrangement of cell states, which significantly increased the accuracy.
“It was interesting for us to see how necessary spatial organization is. Previous studies had shown that cells situated near the mammary duct are necessary. But it’s also necessary to think about which cells are situated near which other cells,” says Zhang.
When they compared the outcomes of their model with samples examined by a pathologist, they found clear agreement in lots of cases. In less clear cases, the model was in a position to provide details about features of a tissue sample, similar to the arrangement of cells, that a pathologist could use in making a call.
This versatile model is also adapted to be used in other varieties of cancer and even neurodegenerative diseases – an area that researchers are also currently investigating.
“We have shown that this easy staining might be very powerful with the appropriate AI techniques. There is far more research to be done, but we want to think about cell organization in additional of our studies,” says Uhler.
This research was funded partially by the Eric and Wendy Schmidt Center on the Broad Institute, ETH Zurich, the Paul Scherrer Institute, the Swiss National Science Foundation, the US National Institutes of Health, the US Office of Naval Research, the MIT Jameel Clinic for Machine Learning and Health, the MIT-IBM Watson AI Lab, and a Simons Investigator Award.
