Why is a diagnosis during surgery so hard to get?
When a surgeon operates on a brain tumour, they need to know quickly what type it is. That information helps decide how much tissue to remove and how to adapt the operation. Today, a small sample is sent to a laboratory, where it is processed, stained and examined under a microscope by a pathologist, the doctor who specialises in reading tissue. This usually takes 20 to 30 minutes.
It also takes staff and equipment, and pathologists are few and unevenly spread between hospitals. The authors point out that this workflow is slow, resource-heavy and depends on a shrinking workforce.
What did the researchers do to speed it up?
The team combined two techniques. The first, called stimulated Raman histology, is a laser-based imaging method. It photographs fresh tissue without staining it or preparing it for long, in just a few minutes.
The second is a convolutional neural network, a type of artificial intelligence built to read images. The researchers trained it on more than 2.5 million such images so it could learn to recognise the main families of brain tumours. They also added a method that highlights the areas of the image where tumour cells have spread into the tissue.
What did the clinical trial find?
The AI gave its diagnosis in the operating room in under 150 seconds, far shorter than the 20 to 30 minutes of the standard method. The approach was then tested in real conditions, in a prospective, multicentre trial: patients were followed as they came, across several hospitals.
In 278 patients, the AI's diagnosis from the laser images was noninferior to the pathologists' reading of conventional slides: 94.6% accuracy for the AI, against 93.9% for the doctors. The authors also note that the network learned visual features that pathologists recognise.
Key figures
- under 150 sTime the AI needed to give a brain tumour diagnosis in the operating room.
- 94.6% vs 93.9%Overall accuracy of the AI compared with pathologists.
- 278 patientsSize of the prospective multicentre clinical trial.
What could this mean for patients?
A fast diagnosis during surgery could help the surgical team make decisions sooner, even in a hospital without a pathologist on site. The authors describe a complementary pathway that does not depend on the usual pathology laboratory.
This does not mean the AI decides in place of the doctor. The final diagnosis is still made by the medical team, with the full tests that follow the operation.
What are the limits of this study?
The study compares the accuracy of the AI with that of pathologists, but it does not show by itself that patients do better because of the tool (for example longer survival or fewer repeat operations). That remains to be proven.
The system was trained to recognise the main families of brain tumours, so rarer or unusual cases could slip past it. The trial also took place in a limited number of centres. Larger studies in other hospitals will be needed before judging its value in routine care.
Brain surgery: feeling supported, with Mima
This research method gives the tumour type during surgery. It is not available in every hospital, and the surgical team decides. The subject is heavy, and it is normal to have questions.
Mima can help you or a loved one note the explanations received, keep track of appointments and explain medical terms in plain words. A question to ask: "Will the tissue be examined during the operation, and when will we get the full results?"
Frequently asked questions
Can AI diagnose a brain tumour during surgery?
According to this 2020 study, yes: the AI gave a diagnosis in under 150 seconds from tumour samples. Across 278 patients, its accuracy (94.6%) was comparable to pathologists (93.9%).
Why does a diagnosis during surgery normally take 20 to 30 minutes?
The sample has to be processed, stained and then read under a microscope by a pathologist. These steps take time and specialist staff.
Does the AI replace the pathologist?
No. The authors present the tool as a complementary pathway. The final diagnosis is still made by doctors.
Is this tool already used in every hospital?
This study is a clinical validation in several centres, not a general roll-out. Its real benefit for patients still has to be shown by further work.
The original study
Hollon TC, Pandian B, Adapa AR, et al. “Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks.” Nature Medicine, 6 January 2020. DOI : 10.1038/s41591-019-0715-9
Read the full study in Nature MedicineThis article summarises a scientific study and is not medical advice.