The Future of Precision Medicine: This AI-enabled tool analyzes brain tumors in real-time during surgery
Brain surgery just got a whole lot more efficient—and precise.
Before operating on a brain tumor, neurosurgeons need to analyze the DNA of their target. Especially when it comes to the most aggressive brain cancers—gliomas—the genetic code holds the secrets of how much tissue should be removed.
Normally, this diagnostic process takes several days—or even weeks. But not anymore.
Harvard Medical School researchers have developed Cryosection Histopathology Assessment and Review Machine (CHARM), an AI tool that detects a brain tumor’s molecular identity very quickly, allowing for real-time analysis during surgery.
According to the team’s paper, published in Med, this tool can also be used outside the operating room to explore tumors on a molecular level, marking a significant leap forward in precision oncology.
Dive in with us as we discuss how this technology works—and what it means for the future of AI-powered precision medicine.
How does CHARM work?
Currently, precision oncology relies on removing brain tissue surgery, freezing it, and then examining it under a microscope. The findings from this process then lead neurosurgeons to decide which and how much brain tissue to remove.
Real-time genetic analysis also allows for otherwise impossible real-time surgical decisions—such as placing targeted chemotherapy “drug wafers” directly into the brain.
“Right now, even state-of-the-art clinical practice cannot profile tumors molecularly during surgery. Our tool overcomes this challenge by extracting thus-far untapped biomedical signals from frozen pathology slides,” said senior study author Kun-Hsing Yu, assistant professor of biomedical informatics at the Blavatnik Institute at HMS.
Yu’s team developed CHARM with the help of 2,334 brain tumor samples from 1,524 people with glioma from three different patient populations. The algorithm was designed to not just analyze these images—but to connect the molecular details with the cells’ appearance, further increasing the model’s accuracy and precision.
If you ask us, where CHARM is most impressive is in that precision.
According to the paper, the tool boasts 93% accuracy in identifying specific molecular mutations—beating the human eye’s ability to discern differences in tissue slides. Specifically, CHARM successfully identified three different types of glioma—all with distinct molecular characteristics, prognoses, and responses to treatment.
It also decreases the need for unnecessary interference in the brain beyond the tissue targeted for removal, which is one of the biggest risks of neurosurgery. Removing too much tissue can affect the patient’s neurocognitive function. On the other hand, removing too little can allow remaining cancerous tissue to spread.
Precision medicine’s next frontier
Of course, CHARM won’t be involved in patient care too soon—it is not yet validated nor FDA approved.
However, the research team is already working on other applications for the technology. Currently, the tool is trained for glioma, but researchers claim it can be retrained to analyze other brain tumor subtypes.
This hits upon an important question when it comes to AI’s application to precision medicine: AI shines when it’s given specific, limited datasets, but will this tool’s predictive ability continue to be as strong when it is retrained on different, perhaps broader datasets?
Similar precision diagnostic AI tools have been developed to analyze malignancies ranging from colon to breast cancer. CHARM is a promising advancement specifically because glioma has been such a challenging target for this kind of technology. It’s an incredibly complex disease with molecular and physical diversity in cell appearance. So, theoretically, if scientists could develop an AI-enabled tool to accurately identify this class of tumors, the odds are high for applications across the wider oncology arena.
However, even to maintain its accuracy for the specific context of gliomas, CHARM will need periodical maintenance and retraining to reflect new disease classifications.
“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.
This brings up another practical question for the future of AI-enabled precision medicine. How much investment will the continued maintenance of these highly-specialized tools require? And will they continue to be as accurate as our knowledge of these diseases develops?
One way to find out? The team behind CHARM has opened the doors for others to test-run the tool for their own projects and datasets. CHARM is available to other researchers for free on GitHub.