Chinese papers appear in international academic journals: demonstrating generative AI applications

[Global Network Technology Comprehensive Report] According to news on September 14, the international academic journal "Nature" magazine "Nature-Communications" recently published an article jointly developed by SenseTime and industry partners, combining generative artificial intelligence and medical care. The latest research results published by Multi-center Federated Learning for Image Data are "Mining Multi-center Heterogeneous Medical Data through Distributed Synthetic Learning".

The paper proposes a federated learning framework DSL based on distributed synthetic adversarial networks, which can use multi-center diverse medical image data to jointly learn the generation of image data. The distributed framework obtains an image data generator through learning, which can generate data more flexibly. These generated data can replace real data from multiple centers for training of specific downstream machine learning tasks, and have strong scalability. .

According to reports, the DSL framework can cleverly solve the common bottleneck of insufficient data volume in medical large model training while protecting data privacy. It can effectively empower MaaS large model training and bring significant benefits to the development iteration of medical large models. breakthrough. With the support of this technology, SenseTime’s “Medical Large Model Factory” can help medical institutions train large medical models for different clinical problems more efficiently and with high quality, further extending the application radius of large models in the medical field

It is worth mentioning that the DSL framework has been verified in multiple specific applications, including brain multi-sequence MRI image generation and downstream brain tumor segmentation tasks, cardiac CTA image generation and downstream whole-heart structure segmentation tasks, and a variety of Pathological image generation of organs and cell nucleus instance segmentation tasks, etc. In terms of scalability, this method can also support different scenarios such as the generation of missing modal data in multi-modal data and continuous learning.

In the exhibition area of ​​Ruijin Hospital, the SenseCare® liver surgery intelligent planning system attracted the attention of many visitors with its efficient and accurate lesion detection, three-dimensional reconstruction and surgical planning functions. It only takes a few minutes to transform a two-dimensional CT image of the liver into a clear three-dimensional model. By gently dragging the mouse, doctors can customize the section, angle, and vessel disconnection position on the model, helping them complete precise liver surgery planning in a few minutes

It is reported that with the launch of the DSL framework, the training of large medical models is expected to break through the shackles of "data islands", lowering the training threshold of large medical models to a certain extent, helping to accelerate model development iterations, making the medical big The application scope of the model can be further extended to cover more clinical medical problems.

The above is the detailed content of Chinese papers appear in international academic journals: demonstrating generative AI applications. For more information, please follow other related articles on the PHP Chinese website!