Deep-learning AI model powers rapid diagnosis for surgeons

Rapid diagnosis during operations helps surgeons determine the scope of work, such as tissue removal in patients with tumors. A new deep learning model leveraging artificial intelligence (AI) may be the best diagnostic tool for these situations, according to a new study conducted by specialists at Brigham and Women’s Hospital (BWH), Boston.

The gold standard process of tissue biopsy and analysis, histology, takes too much time while the patient is on the operating table, and while more recent procedures involving freezing tissue are faster, they also can introduce artifacts that complicate diagnostics. Researchers in the Mahmood Lab at BWH, in conjunction with scientists from Bogazici University, Turkey, developed a better model that uses artificial intelligence to improve image quality and increase rapid diagnosis accuracy.

They published their findings in Natura Biomedical Engineering (6:1407–1419, 2022).

“We are using the power of artificial intelligence to address an age-old problem at the intersection of surgery and pathology,” said corresponding author and pathologist Faisal Mahmood, PhD, associate professor at Harvard Medical School and BWH, in a press release. “Making a rapid diagnosis from frozen tissue samples is challenging and requires specialized training, but this kind of diagnosis is a critical step in caring for patients during surgery.”

Deep learning bridges frozen and gold standard

The gold standard method of biopsy preservation involves formalin-fixed and paraffin-embedded (FFPE) tissue samples, which produces high quality images but can take 12 to 48 hours. Cryosectioning, the fast freezing of tissue followed by cutting and microscopic observation, cuts the turnaround time to mere minutes but can distort cellular details and compromise delicate tissues.

Related article: The seven components of trustworthy artificial intelligence for healthcare

The deep learning method developed by the Mahmood Lab and Turkish colleagues can translate between the histological and cryo methods by subtyping different types of cancer. To test this new method, they conducted a reader study and recruited pathologists to make diagnoses from images created via both the AI and the cryo methods for comparison.

They found the AI method improved image quality and diagnostic accuracy among experts, results reinforced by testing on independently collected data from Turkey.

“Our work shows that AI has the potential to make a time-sensitive, critical diagnosis easier and more accessible to pathologists,” said Mahmood. “And it could potentially be applied to any type of cancer surgery. It opens up many possibilities for improving diagnosis and patient care.”