Scientists have trained AI to diagnose brain tumors on the operating table
[ad_1]
Scientists have trained artificial intelligence to determine the size of a patient’s tumor on the operating table.
This newest tool will tell surgeons how much damaged tissue needs to be removed, scientists from the Netherlands report. Their research was published in the journal Nature, writes The New York Times.
The method involves computer scanning of tumor DNA segments and determination of certain chemical modifications. They can be used to diagnose the type and even subtype of a brain tumor.
“It is critical that the subtype of the tumor is known at the time of surgery. What we have now uniquely enabled is the ability to perform this very fine, reliable, detailed diagnosis already at the time of surgery“, said associate professor of the Center for Molecular Medicine of the Dutch hospital UMC Utrecht and researcher Jeroen de Ridder.
| Photo: shmeljov/Depositphotos |
This AI tool was called Sturgeon. It was first tested on frozen tumor samples from previous surgeries.
Sturgeon accurately diagnosed 45 of 50 cases within 40 minutes of starting genetic sequencing. In another 5 cases, he refrained from offering a diagnosis because the information was unclear to him.
After that, the tool was used during 25 children’s neurosurgery. He determined the correct diagnosis in 18 patients. His research was compared with the standard method of examining tumor samples under a microscope.
In a number of hospitals, it can take weeks to determine the type of tumor. And it takes less than 90 minutes to diagnose Sturgeon. This is fast enough for the surgeon to make a decision before starting the operation, the researchers said.
Dr de Ridder said the model is powerful enough to make a diagnosis using sparse genetic data, similar to how someone recognizes an image based on just one percent of its pixels and from an unknown part of the image.
“He can figure out what he’s looking at and make a reliable classification,” said Dr. de Ridder.
But some tumors are still difficult to diagnose. For example, the samples taken during surgery are about the size of a corn kernel. But if they include some healthy brain tissue, it may be difficult for the AI to pick up enough tumor-specific markers.
Dr. Sebastian Brandner, professor of neuropathology at University College London, noted that sequencing and classifying tumor cells often still requires significant expertise in bioinformatics, as well as staff capable of running, troubleshooting and improving the technology.
Read also: Cancer cells will kill themselves: scientists are developing new drugs against oncology
[ad_2]
Original Source Link
