Harvard-backed AI tool cracks cancer's molecular secrets better than a human

"Our tool overcomes this challenge by extracting thus-far untapped biomedical signals from frozen pathology slides."

Knowing the tumor's molecular identity during surgery is valuable as certain tumors can be treated with drug-coated wafers directly placed into the brain at that moment, according to Yu.

"The ability to determine intraoperative molecular diagnosis in real-time, during surgery, can propel the development of real-time precision oncology," he added.

The current standard intraoperative diagnostic method requires freezing brain tissue and examining it under a microscope. However, freezing can distort cell appearance and impact clinical evaluation accuracy. 

Additionally, despite using powerful microscopes, human eyes struggle to detect subtle genomic variations on slides reliably.

Addressing the glioma gap, CHARM achieves 93% accuracy 

While previous work has seen AI models successfully profile various cancer types, such as colon, lung, and breast, gliomas have posed a greater challenge. This is due to their complex molecular nature and the considerable diversity in the shape and appearance of tumor cells.

To fill this gap, the new AI tool CHARM (Cryosection Histopathology Assessment and Review Machine) was created using 2,334 brain tumor samples from 1,524 glioma patients across three populations. 

It achieved 93 percent accuracy in identifying tumors with specific molecular mutations and successfully classified three major glioma types with distinct features and varied prognoses and treatment responses. 

CHARM also accurately captured visual characteristics, spotting regions with higher cellular density and increased cell death, indicating aggressive glioma types.

According to Yu, CHARM can look at the bigger picture around an image, like how a human pathologist would visually check out a tumor sample.

While trained and tested on glioma samples, the researchers suggest that the model can be successfully retrained to identify other subtypes of brain cancer.

"Just like human clinicians who must engage in ongoing education and training, AI tools must keep up with the latest knowledge to remain at peak performance," Yu said. 

Although CHARM is currently accessible to other researchers, it must undergo real-world testing and receive FDA clearance before being deployed in hospitals for clinical validation.