Why does MSI matter for immunotherapy?
Immunotherapy helps the body's defense system attack cancer. It has changed care for several cancers, such as melanoma and lung cancer. In digestive cancers it usually works less well, except for one group of tumors with microsatellite instability, or MSI. In these tumors, the system that repairs DNA errors does not work properly.
About 15% of bowel (colorectal) and stomach cancers are MSI. For these patients, immunotherapy drugs have shown real benefit and have been approved by the US medicines regulator, the FDA. But finding out whether a tumor is MSI needs an extra test, either genetic or by staining the tissue in a special way. According to the authors, not every patient is tested, except in large specialist centers.
What did the researchers do?
The team used slides stained with hematoxylin and eosin, or H&E: the pink and purple stain used for almost every biopsy. A first deep learning AI finds the tumor areas on the scanned image. These areas are cut into small squares, and a second AI estimates for each square whether it looks like an MSI tumor.
The models were trained on hundreds of patients from the public TCGA database, with bowel or stomach cancer. They were then tested on patients they had never seen: 378 patients from an independent study called DACHS, 185 Japanese patients with stomach cancer, and patients with cancer of the womb lining (endometrial cancer).
What were the results?
Results are given as the area under the curve, a score between 0.5 (chance) and 1 (a perfect test). In patients held back from the TCGA database, the AI scored from 0.77 to 0.84 in bowel cancer and 0.81 in stomach cancer. In the 378 patients of the external DACHS group, it reached 0.84.
Performance fell when the population changed: 0.69 in Japanese patients with stomach cancer, a disease that looks different in this group, and 0.75 in endometrial cancer. The researchers also showed that a small amount of tissue, similar to a biopsy, was enough for a stable result.
Key figures
- 15%approximately, of bowel and stomach cancers are MSI
- 0.84area under the curve in 378 patients from an external group (DACHS)
- 0.69in 185 Japanese patients with stomach cancer
Talking about your MSI status with your team, with Mima
This AI is still research: today, MSI status is checked with a genetic test or a tissue stain. But the study is a reminder of how much this result can matter for access to immunotherapy.
Mima helps you keep track of your reports and treatments and prepare your questions for appointments, such as: "Has my tumor been tested for MSI?" She also explains complicated words in plain language.
What could this mean for patients?
Stained slides exist for almost every cancer patient. If an AI could use them to flag tumors that are likely to be MSI, more patients could be tested and then offered immunotherapy. The authors estimate that the method could be set up at low cost in large centers that already scan their slides.
For a patient, this points to a question for the care team: has my tumor been tested for MSI?
What are the limits of this study?
This is a first demonstration, carried out on existing slides. The authors say the AI needs training on larger datasets and prospective validation before it can be used in care. The standard MSI test is therefore still needed.
Results dropped when the population differed from the one used for training, as with the Japanese patients. Finally, the study does not yet show that any patient received better care thanks to this tool.
Frequently asked questions
What does MSI mean?
MSI stands for microsatellite instability. It is a feature of some tumors whose DNA is poorly repaired, which often makes them responsive to immunotherapy.
How is MSI tested today?
With a genetic test or a special tissue stain called immunohistochemistry. The AI in this study uses only the usual stained slide.
Does this AI replace the MSI test?
No. The authors describe a method that still needs prospective studies. The standard test is still needed.
Which cancers does it concern?
The study mainly covers bowel and stomach cancers, with a test on endometrial cancer.
The original study
Kather JN, Pearson AT, Halama N, et al. “Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer” Nature Medicine, June 3, 2019. DOI : 10.1038/s41591-019-0462-y
Read the full study in Nature MedicineThis article summarizes a scientific study and is not medical advice.