"We are using the power of artificial intelligence to address an age-old problem at the intersection of surgery and pathology," said corresponding author Faisal Mahmood, PhD, of the Division of Computational Pathology at BWH. "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."
For making final diagnoses, pathologists use formalin-fixed and paraffin-embedded (FFPE) tissue samples - this method preserves tissue in a way that produces high-quality images but the process is laborious and typically takes 12 to 48 hours. For a rapid diagnosis, pathologists use an approach known as cryosectioning that involves fast freezing tissue, cutting sections, and observing these thin slices under a microscope. Cryosectioning takes minutes rather than hours but can distort cellular details and compromise or tear delicate tissue.
Mahmood and co-authors developed a deep-learning model that can be used to translate between frozen sections and more commonly used FFPE tissue. In their paper, the team demonstrated that the method could be used to subtype different kinds of cancer, including glioma and non-small-cell lung cancer. The team validated their findings by recruiting pathologists to a reader study in which they were asked to make a diagnosis from images that had gone through the AI method and traditional cryosectioning images. The AI method not only improved image quality but also improved diagnostic accuracy among experts. The algorithm was also tested on independently collected data from Turkey.
The authors note that in the future, prospective clinical studies should be conducted to validate the AI method and determine if it can contribute to diagnostic accuracy and surgical decision-making in real hospital settings.
"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."
Ozyoruk KB, Can S, Darbaz B, Başak K, Demir D, Gokceler GI, Serin G, Hacisalihoglu UP, Kurtuluş E, Lu MY, Chen TY, Williamson DFK, Yılmaz F, Mahmood F, Turan M.
A deep-learning model for transforming the style of tissue images from cryosectioned to formalin-fixed and paraffin-embedded.
Nat Biomed Eng. 2022 Dec;6(12):1407-1419. doi: 10.1038/s41551-022-00952-9