Game-changing AI tool makes real-time diagnoses during surgery

When surgeons send samples to a pathologist for examination, both speed and accuracy are important. (CREDIT: Creative Commons)

When a patient undergoes surgery to remove a tumor or treat a disease, the course of the operation is often not predetermined. To decide how much tissue to remove, surgeons must know more about the condition they are treating, including the margins of the tumor, its stage, and whether the lesion is malignant or benign—determinations that often depend on collection, analysis, and diagnosis. diseases. while the patient is on the operating table.

When surgeons send samples to a pathologist for examination, both speed and accuracy are important. The current gold standard for tissue examination often takes too long, and a faster approach involving tissue freezing can introduce artifacts that can make diagnosis difficult.

A new study led by researchers from Mahmoud’s lab at Brigham and Women’s Hospital, co-founder of the Mass General Brigham healthcare system, and collaborators at Bogazici University has come up with a better way; the method uses artificial intelligence to translate between frozen sections and the gold standard approach, improving image quality to increase the accuracy of rapid diagnosis.

The results are published in the journal Nature Biomedical Engineering.

Similar stories

“We are harnessing the power of artificial intelligence to solve an age-old problem at the intersection of surgery and pathology,” said corresponding author Faisal Mahmood, PhD, from the Computational Pathology Department at BWH. “Rapid diagnosis from frozen tissue samples is challenging and requires specialized training, but such diagnosis is an important step in patient care during surgery.”

Formalin-fixed, paraffin-embedded (FFPE) tissue samples are used by pathologists to make a definitive diagnosis. This method allows tissue to be preserved in such a way as to produce high quality images, but the process is labor intensive and usually takes 12 to 48 hours.

For rapid diagnosis, pathologists use an approach known as cryosectioning, which involves rapidly freezing tissue, cutting the sections, and observing these thin sections under a microscope. Cryosection takes minutes, not hours, but can distort cellular detail and damage or tear delicate tissue.

Workflow for transferring a cryosection to FFPE. (CREDIT: Nature Biomedical Engineering)

Mahmood et al have developed a deep learning model that can be used to translate between frozen sections and the more commonly used FFPE tissue.

In their paper, the team demonstrated that this technique can be used to subtype a variety of cancers, including glioma and non-small cell lung cancer. The team backed up their findings by recruiting pathologists to participate in a study that asked them to make a diagnosis based on AI images and traditional cryosection images.

Architecture of the AI-FFPE method. (CREDIT: Nature Biomedical Engineering)

The AI ​​method has not only improved the image quality, but also increased the diagnostic accuracy among experts. The algorithm has also been tested on independently collected data from Turkey.

The authors note that prospective clinical studies should be conducted in the future to test the AI ​​method and determine whether it can contribute to accurate diagnosis and surgical decision making in a real hospital setting.

“Our work shows that AI can simplify and make critical time-based diagnosis easier and more accessible for pathologists,” Mahmood said. “And it could potentially be applied to any type of cancer surgery. This opens up many opportunities to improve diagnosis and patient care.”

For more science news, visit our New Innovations section at The bright side of the news.

Note. Materials provided by Brigham and Women’s Hospital above. Content can be edited for style and length.

Do you like such pleasant stories? Receive Brighter Side of News Newsletter.