A new AI model developed by researchers at Harvard Medical School is making waves in cancer diagnosis and treatment. This AI model can perform a variety of tasks for 19 different types of cancer. It goes beyond many current AI systems that are usually limited to one specific job, like just detecting cancer or predicting a tumor’s genetic makeup.
The AI, named CHIEF (Clinical Histopathology Imaging Evaluation Foundation), analyzes digital images of tumor tissues. It detects cancer cells, predicts the tumor’s molecular profile, and even forecasts patient survival rates. Researchers found that CHIEF is more accurate than most current AI methods, achieving nearly 94% accuracy in detecting cancer.
The researchers believe CHIEF could help identify patients who might benefit from experimental treatments that target specific genetic variations. This capability is especially valuable in areas with limited access to advanced testing.
The model was trained on a large dataset of 15 million images and then refined with 60,000 whole-slide images from various types of cancer, including lung, breast, and colon. CHIEF’s ability to analyze both specific parts and the whole image allows it to understand the context better, making it more effective.
In tests, CHIEF outperformed other AI methods by up to 36% in tasks like detecting cancer, identifying tumor origins, and predicting patient outcomes. It achieved nearly 94% accuracy in cancer detection and 96% accuracy in specific biopsy datasets.
CHIEF also predicts genetic variations that influence how a tumor will respond to treatments. This is important because current methods of genetic testing can be time-consuming and expensive. By identifying specific patterns in tissue images, CHIEF offers a faster, more cost-effective alternative.
Additionally, CHIEF can predict patient survival based on initial tumor images, helping doctors distinguish between patients likely to survive longer versus those at higher risk. This ability has been validated across multiple patient groups and institutions.
The team plans to improve CHIEF by training it on images of rare diseases, pre-cancerous tissues, and more molecular data to help identify aggressive cancers. They also aim to enhance its ability to predict the effects of new cancer treatments.