No abstract

Artificial intelligence provides a promising solution for streamlining COVID-19 diagnoses; however, concerns surrounding security and trustworthiness impede the collection of large-scale representative medical data, posing a considerable challenge for training a well-generalized model in clinical practices. To address this, we launch the Unified CT-COVID AI Diagnostic Initiative (UCADI), where the artificial intelligence (AI) model can be distributedly trained and independently executed at each host institution under a federated learning framework without data sharing. Here we show that our federated learning framework model considerably outperformed all of the local models (with a test sensitivity/specificity of 0.973/0.951 in China and 0.730/0.942 in the United Kingdom), achieving comparable performance with a panel of professional radiologists. We further evaluated the model on the hold-out (collected from another two hospitals without the federated learning framework) and heterogeneous (acquired with contrast materials) data, provided visual explanations for decisions made by the model, and analysed the trade-offs between the model performance and the communication costs in the federated training process. Our study is based on 9,573 chest computed tomography scans from 3,336 patients collected from 23 hospitals located in China and the United Kingdom. Collectively, our work advanced the prospects of utilizing federated learning for privacy-preserving AI in digital health.

This study combined measurements from the Chinese operational geostationary satellite Fengyun‐2E (FY‐2E) and ground‐based weather radars to conduct a statistical survey of isolated convection initiation (CI) over central eastern China (CEC). The convective environment in CEC is modulated by the complex topography and monsoon climate. From May to August 2010, a total of 1,630 isolated CI signals were derived from FY‐2E using a semiautomated method. The formation of these satellite‐derived CI signals peaks in the early afternoon and occurs with high frequency in areas with remarkable terrain inhomogeneity (e.g., mountain, water, and mountain‐water areas). The high signal frequency areas shift from northwest CEC (dry, high altitude) in early summer to southeast CEC (humid, low altitude) in midsummer along with an increasing monthly mean frequency. The satellite‐derived CI signals tend to have longer lead times (the time difference between satellite‐derived signal formation and radar‐based CI) in the late morning and afternoon than in the early morning and night. During the early morning and night, the distinction between cloud top signatures and background terrestrial radiation becomes less apparent, resulting in delayed identification of the signals and thus short and even negative lead times. A decline in the lead time is observed from May to August, likely due to the increasing cloud growth rate and warm‐rain processes. Results show increasing lead times with increasing landscape elevation, likely due to more warm‐rain processes over the coastal sea and plain, along with a decreasing cloud growth rate from hill and mountain to the plateau.

No abstract