Harvard Scientists Develop AI Model with 96% Cancer Detection Accuracy

Researchers at Harvard Medical School have developed an AI model named Chief, capable of accurately identifying various types of cancer, assessing treatment methods, and predicting survival rates. This is detailed in an article on Nature.

According to the scientists, their neural network is groundbreaking due to the wide range of tumors it can analyze and its ability to predict patient outcomes.

Artificial intelligence has enhanced image-based diagnostic methods by identifying features that even experienced individuals might overlook.

“We aimed to create a flexible and versatile AI platform, similar to ChatGPT, capable of performing a wide range of cancer assessment tasks. Our model has proven extremely useful in tumor detection and predicting treatment responses across various diseases,” commented Kun-Hsing Yu, Associate Professor of Biomedical Informatics at the Blavatnik Institute, Harvard Medical School.

Chief evaluates digital images of tumor tissues. It is trained on 15 million unlabeled fragments and 60,000 full-sized images, covering 19 different types of cancer.

The model reportedly outperformed other AI diagnostic methods by 36% in detecting cancer cells, predicting disease outcomes, determining tumor origins, and identifying genetic patterns related to treatment responses.

Chief detects cancer with 94% accuracy. This figure increased to 96% following analysis of the esophagus, stomach, colon, and prostate. The model provides additional information about the tissues surrounding the tumor, including the presence of more immune cells in individuals who have lived with cancer for a long time compared to those who died early.

Yu believes that if Chief and similar approaches continue to be explored, they could be used in the future for “early identification of cancer patients who might benefit from experimental treatments targeting specific molecular variations.”

Previously, the AI model EMethylNET was able to detect 13 different types of cancer with 98.2% accuracy based on DNA data from tissue samples.

In July, scientists taught artificial intelligence to diagnose Alzheimer’s disease with 70% accuracy seven years before its onset and 80% a year prior.