AI may be able to help in the fight against cancer.
Kun-Hsing Yu, associate professor in the department of biomedical informatics at Harvard Medical School, looks into how it can help.
Kun-Hsing “Kun” Yu, M.D., Ph.D., is an Associate Professor in the Department of Biomedical Informatics at Harvard Medical School. He pioneered the first fully automated artificial intelligence (AI) algorithm capable of extracting thousands of features from whole-slide pathology images. His research has uncovered molecular mechanisms driving the microscopic phenotypes of tumor cells and identified novel cellular morphologies that predict patient prognosis.
Dr. Yu’s lab integrates multi-omics (e.g., genomics, epigenomics, transcriptomics, and proteomics) profiles with quantitative pathology patterns to predict clinical phenotypes in cancer patients. The AI methods developed by the Yu Lab have been independently validated by over 80 research laboratories worldwide.
His contributions to AI in pathology have earned numerous honors, including the National Institutes of Health (NIH) Maximizing Investigators’ Research Award, Google Research Scholar Award, American Medical Informatics Association New Investigator Award, Harvard Medical School Dean’s Innovation Award, Department of Defense (DoD) Career Development Award, and the American Cancer Society (ACS) Research Scholar Award. He is a Fellow of the American Medical Informatics Association (FAMIA).
AI Distinguishes Glioblastoma from Look-Alike Cancers
Every year, more than 300,000 people worldwide are diagnosed with cancers that arise in the brain. One of the most devastating is glioblastoma, an aggressive tumor that typically requires immediate surgery. But there’s a challenge: on clinical evaluation and imaging tests, glioblastoma can look nearly identical to another cancer called primary central nervous system lymphoma.
When it comes to treatment, the difference matters. Unlike glioblastoma, lymphoma in the brain should not be surgically removed — it responds best to chemotherapy and radiation. Confusing one for the other can have serious consequences.
To address this, our team developed an AI tool called PICTURE — short for Pathology Image Characterization Tool with Uncertainty-aware Rapid Evaluations — that analyzes digital images of tumor samples taken during surgery. We tested PICTURE across five hospitals in four countries and found it could distinguish glioblastoma from lymphoma with more than 98 percent accuracy, outperforming both human pathologists and other AI models.
What makes PICTURE especially powerful is its ability to say, “I’m not sure.” If it encounters a tumor type it hasn’t seen before, it flags the case for human review. This is especially important in brain cancer, which includes over 100 tumor types—some extremely rare. By recognizing its own limits, PICTURE reduces the risk of a dangerous misdiagnosis.
We see PICTURE as a partner to surgeons and pathologists in the operating room, where speed and accuracy can determine a patient’s outcome. By minimizing diagnostic errors, it helps ensure patients receive the right treatments at the right time.
Looking ahead, we hope PICTURE will bring advanced cancer diagnostics to patients worldwide—and serve as a teaching tool for the next generation of doctors, supporting rather than replacing experts.
Read More:
[Nature] – Uncertainty-aware ensemble of foundation models differentiates glioblastoma from its mimics
