Yandex, ISP RAS and Sechenov University: Neural network for detecting atrial fibrillation

The main articles are:

• Neural networks (neural networks)
• Artificial Intelligence in Medicine
• Machine learning
• Diagnosing Heart Disease

2024: Building a neural network to detect atrial fibrillation

Yandex will allow partners to train neural networks together and store data separately. The company announced this on October 8, 2024.

This opportunity is given by federated machine learning - Yandex, together with the ISP RAS and Sechenov University, tested it in practice for medical problems. A federated approach allows organizations, such as banks or medical institutions, to participate in partnership projects without transferring their sensitive data to the outside.

A federated approach allows participants to collectively train models without passing on their data to anyone. This opens up new opportunities for partnerships in the field of artificial intelligence. Now companies from industries dealing with sensitive information will be able to participate in them: for example, finance, medicine or industry.

First, the model is trained at each of the datasets of the project participants. It is not the datacets themselves that are transmitted to the central server, but the results of their processing - for example, changes in the model weights. It is on them that the global model is then trained. Thus, datacets throughout the training process do not leave the contours of organizations - and no one except the owners can access them.

Yandex, ISP RAS and Sechenov University of the Ministry of Health of Russia, using a federal approach, created a neural network that, according to electrocardiograms, detects atrial fibrillation - one of the most common heart pathologies. The technology does this with high rates of sensitivity and specificity. For training, two independent datasets with electrocardiograms were used: from Sechenov University and from the ISP RAS. Both partners conducted rounds of training on their side, and then transferred the results to the general circuit.

The project was implemented by experts from the technology center for the Yandex Cloud society. Yandex Cloud and ISP RAS engineers were responsible for the technical component of the project. Yandex Cloud thought through the implementation stages, proposed a technology stack, created a unified learning environment, and calculated the required amount of resources. The ISP RAS developed a model and adapted it to the open source framework of federated learning. Sechenov University gave an expert assessment of the model quality.

In the future, Yandex Cloud customers will be able to use federated machine learning. This approach will allow organizations to participate in joint projects that could not cooperate before due to the risks associated with the transfer of sensitive data. This, in turn, will increase the quality of the final models - the more partners in the project, the more data for training. In addition, the federated method will also be useful for partners who are separated by long distances - for example, when it comes to cross-border data transfer.