One of the most aggressive forms of cancer could be detected at an early stage using an AI tool

One of the most aggressive forms of cancer could be detected at an early stage using an AI tool

A team of Chinese researchers has developed an artificial intelligence (AI) platform that can help detect pancreatic cancer, one of the most aggressive forms of cancer, at an early stage based on imaging at high-precision X-rays, Xinhua reported on Wednesday. by Agerpres.

Pancreatic tissue sample under microscopePhoto: David Davies / PA Images / Profimedia

PANDA, a “deep learning” method developed by Alibaba’s Damo Academy, can amplify and identify certain subtle pathological features in simple CT images that are difficult to detect with the naked eye.

Pancreatic cancer is a profoundly malignant tumor, but it can be cured if detected early.

Initially, the PANDA tool was trained using a dataset from a total of 3,208 patients from a single center. According to a study published this week in the scientific journal Nature Medicine, the AI ​​tool achieved a sensitivity of 94.9% and a specificity of 100% in an internal testing group of 291 patients at the Pancreas Institute of Shanghai.

The model was then validated on external groups consisting of people from several centers – in total, 5,337 patients from China and the Czech Republic – based on abdominal CT scans.

Its sensitivity and specificity were 93.3% and 98.8%, respectively, surpassing the performance of an average radiologist, Xinhua noted. Sensitivity refers to the ability of a test to classify an individual with the disease as positive, while the specificity of a test refers to its ability to classify an individual without the disease as negative.

The PANDA model has been used in clinical settings more than 500,000 times, recording only one false positive result out of a thousand tests. PANDA could become a new widespread screening tool for pancreatic cancer, the researchers said.



Leave a Reply

Your email address will not be published. Required fields are marked *