South SFV Today

Researchers at Stanford Medicine have developed an artificial intelligence model, named MUSK, that can integrate both visual and language-based information to predict cancer prognoses and treatment responses. The model was trained on a vast dataset of 50 million medical images and over 1 billion pathology-related texts. According to Ruijiang Li, MD, an associate professor of radiation oncology and senior author of the study published in Nature on January 8th, "MUSK can accurately predict the prognoses of people with many different kinds and stages of cancer."

The model has shown promise in identifying patients with lung or gastroesophageal cancers who may benefit from immunotherapy and pinpointing melanoma patients likely to experience recurrence. Postdoctoral scholars Jinxi Xiang, PhD, and Xiyue Wang, PhD are lead authors of the research.

Unlike traditional AI models used primarily for diagnostics, MUSK serves as a foundation model capable of using unpaired multimodal data. This allows it to leverage more extensive datasets for training compared to conventional methods. "We designed MUSK because, in clinical practice, physicians never rely on just one type of data to make clinical decisions," Li explained.

MUSK demonstrated its effectiveness by predicting disease-specific survival rates with higher accuracy than standard predictions based on cancer stage alone. It also outperformed existing methods in determining which non-small cell lung cancer patients would benefit from immunotherapy treatments.

The development team included contributions from researchers at Harvard Medical School. Funding was provided by several grants from the National Institutes of Health and support from the Stanford Institute for Human-Centered Artificial Intelligence.