Please click here to download the model.
Abstract:
Coronavirus disease 2019 (COVID-19) has caused considerable morbidity and mortality, especially in patients with underlying health conditions. A precise prognostic tool to identify poor outcomes among such population is desperately needed.
Total 400 COVID-19 patients with underlying health conditions were retrospectively recruited from 4 centers, including 54 dead cases (labeled as poor outcomes) and 346 patients discharged or hospitalized for at least 7 days since initial CT scan.
We first developed an initial CT-derived deep learning model based on 3D-ResNet10 to predict the probability of COVID-19 patients reaching poor outcome. Users can test your data using our deep learning model code and the model weights (Python 3.6, PyTorch 1.3.0).
For test input, we adopted an automatic lung volume segmentation scheme via the threshold segmentation and flood fill algorithm. Then, the segmented lung volume was resized to a 3D volume of interest (VOI) with the size of 48×240×360, and the VOIs were normalized by the min-max normalization method to minimize the influence of voxel distribution and contrast variation. Finally, 24 slices were center cropped and used as the input.
If you get a deep learning model probability > 0.1714, the patient is more likely to reach poor outcome (death).