New AI technology improves accuracy of COVID-19 diagnosis

An international team led by professionals from University of Cambridge in the United Kingdom and Huazhong University of Science and Technology in China used A technique called Federal Learning to build a new model that allows detecting the presence of the COVID-19 virus through artificial intelligence (AI) practices.With Federal Learning, an AI model can be applied in a hospital or a country, and you can verify independently using a dataset from another hospital or country, without sharing information.

The researchers based their model on more than 9,000 CT scans of nearly 3,300 patients in 23 hospitals in the UK and China. Your search resultsPublished in the magazine The intelligence of nature’s machineAnd They provide a framework in which AI technologies can be more reliable and accurate, especially in areas such as medical diagnosis, where privacy is vital.

Diagnostics through artificial intelligence have provided a promising solution to accelerate the outcome of COVID-19 and future public health crises. but nevertheless, Safety and reliability concerns prevent the large-scale collection of representative medical data, Which poses a challenge to train a model that can be used all over the world.

During the early days of the pandemic, many AI researchers worked to develop models that could diagnose the disease. However, many of them were generated using low-quality data, incomplete or incomplete data sets, and a lack of information from clinicians. Several researchers in the current study note that these older models were not suitable for clinical use during the 2021 pandemic season.

“Artificial intelligence has many limitations when it comes to diagnosing COVID-19, and we must carefully analyze and select the data until we end up with a model that works and is reliable.”First co-author Hanshin Wang of Cambridge, of the University of Cambridge’s Department of Engineering, explained. And added: “When previous models were based on arbitrary open source data, we worked with a large team of NHS radiologists and Wuhan Tongji Hospital Group to identify the data, so we started from a strong position.”.

The researchers used two well-chosen, appropriately sized sets of external validation information to test their model and ensure that it would work well on data sets from different hospitals or countries. “Before COVID-19, people didn’t realize how much data they needed to collect to create medical AI applications,” noted co-author Michael Roberts of AstraZeneca, responsible for Cambridge’s Department of Applied Mathematics and Theoretical Physics.

“Different hospitals, different countries, they all have their own ways of doing things, so you need the data sets to be as large as possible to have useful tracking of the widest range of physicians.”

The researchers based their framework on 3D computed tomography rather than 2D images. CT scans provide a much higher level of detail, which leads to a better model.They used 9,573 CT scans from 3,336 patients collected from 23 hospitals located in China and the United Kingdom.

What is more, They had to mitigate bias caused by different datasets and used standardized learning to train a better generalized AI model, while keeping each data center private in a collaborative environment. For a fair comparison, The researchers validated all models with the same data, without interfering with the training data. The team had a panel of radiologists who made diagnostic predictions based on the same set of CT scans and compared the accuracy of the AI ​​models to that obtained by the specialists.

The researchers say their model is useful not only for COVID-19, but for any other disease that can be diagnosed with a CT scan. “Next time there is a pandemic, and there is every reason to believe there will be, we will be in a much better position to take advantage of AI technologies quickly so that we can understand new diseases more quickly,” Wang said.

“We’ve shown that medical data can be encrypted, so we can create and use these tools while maintaining patient privacy across internal and external borders. By working with other countries, we can do much more than we can on our own,” Roberts added.

Researchers are currently collaborating with the newly established WHO Center for Epidemiological Intelligence and Epidemiology to explore the potential to enhance privacy-preserving digital healthcare frameworks.

Read on: