An artificial intelligence developed by Harvard Medical School can diagnose a variety of cancers with almost 94 percent accuracy, its creators say. Named CHIEF (Clinical Histopathology Imaging Evaluation Foundation), researchers say it was tested on 19 types of cancer, setting it apart from other models.
According to the scientists, CHIEF is much more flexible than other medical AIs and has the ability to perform a wider range of tasks. As well as diagnosing cancer, CHIEF can predict how a patient might react to a certain treatment.
“Our ambition was to create a nimble, versatile ChatGPT-like AI platform that can perform a broad range of cancer evaluation tasks,” said Kun-Hsing Yu, assistant professor of biomedical informatics at Harvard Medical School. “Our model turned out to be very useful across multiple tasks related to cancer detection, prognosis, and treatment response across multiple cancers.”
CHIEF was trained on 15 million unlabeled images. That was followed by 60,000 images of tissues, to allow CHIEF to learn the broader context of cancers and their development. After that training, CHIEF was tested on more than 19,000 images from 32 global datasets.
Harvard said that CHIEF outperformed other comparable AIs by up to 36 percent in tasks that included detecting cancer cells, identifying tumor origins, predicting outcomes, and identifying DNA patterns. CHIEF’s work could cut out the need for expensive and time-consuming DNA sequencing to suggest the best treatments for some patients.
CHIEF’s overall detection accuracy is around 94 percent and rises to 96 percent for specific types of cancers, researchers say. This versatility means CHIEF can perform accurately whether the cells were obtained via biopsy or surgical excision, they added.
The team says the plan is to train CHIEF even further on images of rare and non-cancerous disease, as well as pre-cancerous cells. They also plan to train it to predict the effects of novel cancer treatments on top of regular ones.
The post Harvard Medical School’s new AI can diagnose cancer with very high accuracy appeared first on ReadWrite.
