※ HUST-19
1. The HUST-19 1.0 is released on August 18, 2020:
    For training the individual image-based model in HUST-19, we manually labelled 19,685 CT slices, including 5705 non-informative CT (NiCT), 4001 positive CT (pCT) and 9979 negative CT (nCT) slices, randomly selected from 61 and 43 patients with and without COVID-19 pneumonia.
    For training models to predict morbidity outcomes, we used 197,068 CT slices and 127 types of CF data from 222 control, 438 Type I, and 211 Type II patients in the Cohort 1. We use the Cohort 2 as an independent dataset to test HUST-19, which still exhibits a promising accuracy. In the Cohort 2, there are 91,430 CT slices and CF data from 106 controls, 182 Type I patients, and 63 Type II patients. For training models to predict mortality outcomes, the Cohort 1 and 2 are merged with 169,933 CT slices and CF data from 662 cured and 57 deceased cases, due to data limitation. All computational models of HUST-19 are made available under a CC BY-NC 4.0 license.

2. Inception Net V3 and ChexNet:
    Besides HUST-19, we also implemented two additional open-source CNN frameworks, Inception Net V3 (Szegedy, C., et al., 2016) and ChexNet (Rajpurkar, P., et al., 2017), for predicting morbidity or mortality outcomes using our CT data, respectively. Inception Net V3 and ChexNet models are also made available under a CC BY-NC 4.0 license.