Federated learning of a computer vision system for automated interpretation of medical images

训练用于自动图像解释的计算机视觉算法需要一个中央数据存储库。当数据包括医学成像检查时，需要从不同的医院池中提取数百万条历史记录，并将数据传输到中央数据中心，这是一项不是微不足道的任务，目前也是构建放射学人工智能系统时需要克服的最大障碍之一。例如，数据的敏感性需要在每个站点进行大规模匿名化，这是一个非常昂贵和耗时的过程。在这个项目中，我们将为联合学习开发一种新的机器学习方法，使位于各医院的分散放射存储库能够协作学习计算机视觉的共享机器学习模型，同时将所有训练数据保留在现场，将进行机器学习的能力与将数据存储在中央位置的需求分离。联合学习系统将与NHS医院合作开发和测试，将允许更智能的模型、更低的延迟和更少的功耗，所有这些都将确保隐私。我们的方法将简化AI系统的交付，以支持跨成像模式(例如X射线、CT和MRI)的放射报告。

Training computer vision algorithms for automated image interpretation requires a centralised data repository. When the data consists of medical imaging examinations, the need to extract millions of historical records from a pool of different hospitals, and transfer the data into a centralised data centre, is a non-trivial task and currently one of the largest hurdles to overcome when building AI systems in radiology. For instance, the sensitivity of the data requires large-scale anonymization at each site, which is a very costly and time consuming process.In this project we will develop a new machine learning approach for Federated Learning enabling decentralised radiological repositories located at each hospital to collaboratively learn a shared machine learning model for computer vision while keeping all the training data on site, decoupling the ability to do machine learning from the need to store the data in a centralised location.The Federated Learning system will be developed and tested in collaboration with NHS hospitals, and will allow for smarter models, lower latency, and less power consumption, all while ensuring privacy. Our approach will simplify the delivery of AI systems to support radiological reporting across imaging modalities (e.g. X-ray, CT and MRI).Alligns with the artificial intelligence technologies research area.