Tuesday, February 27, 2024
HomeTechnologyAI uses medical notes to teach itself to spot diseases on chest...

AI uses medical notes to teach itself to spot diseases on chest X-rays

After processing thousands of chest X-rays and accompanying clinical reports, the AI ​​learned to spot disease in those scans as accurately as a human radiologist.

Current diagnostic AI models are trained on scans labeled by humans, but labeling is a time-consuming process. The new model, called CheXzero, can “learn” on its own from existing medical reports written by experts in natural language.

Findings show that labelling X-rays is unnecessary for training AI models to interpret medical images, which can save time and money.

A team of researchers at Harvard Medical School trained the CheXzero model on a publicly available dataset of over 377,000 chest X-rays and over 227,000 corresponding clinical reports. This taught it to associate certain types of images with its existing notes, rather than learning from structured data manually labeled for the task.

The performance of CheXzero was subsequently tested on different datasets from two different institutions, one in the other country, to check whether it was able to match images with the corresponding annotations, even if the report contained different terms.

The study, described in Nature Biomedical Engineering, found that the model outperformed other self-supervised AI models are more effective. In fact, it is similar to human radiologists in accuracy.

While others have attempted to use unstructured medical data in this way, this is the first time the team’s AI model has learned from unstructured text and matched the report’s co-authors, Stanford undergraduate student and visiting researcher Ekin Tiu said radiologists performed and demonstrated that it could predict multiple diseases with high accuracy from a given X-ray.

“We were the first to do this and demonstrate it effectively in the field,” he said.

The code for the model has been made public to other researchers in the hope that it can be used as a tool for the technology application, said Pranav Rajpurkar, assistant professor of biomedical informatics at the Blavatnik Institute at Harvard Medical School, who led the project. on CT scans, MRIs, and echocardiograms to help detect broader disease in other parts of the body.

“We want people to be able to apply it to other chest X-ray datasets and diseases that they care about,” he said.

Rajpurkar is also optimistic that diagnostic AI models that require minimal oversight can help increase access to healthcare in countries and communities where experts are scarce.

“It makes a lot of sense to use richer training signals from reports,” said Christian Leibig, head of machine learning at German startup Vara, which uses artificial intelligence to detect breasts cancer. “Achieving this level of performance is quite an achievement.”



Please enter your comment!
Please enter your name here


Featured NEWS