New Artificial Intelligence at Harvard Medical School Achieves Cancer Diagnosis

2024-09-06

Harvard Medical School has developed an artificial intelligence (AI) that can diagnose multiple types of cancer with an accuracy rate of nearly 94%. The AI, called CHIEF (Clinical Histopathological Image Evaluation Foundation), has been tested on 19 different types of cancer, setting it apart from other models.

What can Harvard's AI do?

Scientists say that CHIEF is more flexible than other medical AIs and can perform a wider range of tasks. In addition to diagnosing cancer, CHIEF can also predict a patient's response to specific treatments.

Kun-Hsing Yu, Assistant Professor of Biomedical Informatics at Harvard Medical School, said, "Our goal is to create an agile, versatile ChatGPT-style AI platform that can perform a wide range of cancer assessment tasks. Our model is very useful in multiple tasks related to cancer detection, prognosis, and treatment response, covering various types of cancer."

CHIEF was trained on 15 million unlabeled images. It then learned from 60,000 tissue images to gain a broader understanding of cancer and its development. After training, CHIEF was tested on over 19,000 images from 32 global datasets.

Harvard University states that CHIEF outperforms other comparable AIs by up to 36% in tasks such as detecting cancer cells, identifying tumor origins, predicting outcomes, and recognizing DNA patterns. The work of CHIEF could reduce the need for expensive and time-consuming DNA sequencing that some patients undergo to obtain the best treatment plan.

Researchers report that CHIEF has an overall detection accuracy of approximately 94%, which increases to 96% for certain specific types of cancer. This flexibility means that CHIEF can accurately detect cells whether they are obtained through biopsies or surgical removal.

The research team plans to further train CHIEF on images of rare diseases, non-cancerous conditions, and pre-cancerous cells. Additionally, they plan to train CHIEF to predict the effectiveness of novel cancer treatments based on conventional treatment methods.