| dc.description.abstract | Background: The data of the first episode of the COVID-19 pandemic in Taiwan is scarce. We researched the risk factors of death among mechanically-ventilated patients with COVID-19 in Taiwan during the first episode of COVID-19. In addition, we are inspired to create a new artificial-intelligence-based death prognostication model by utilization of chest X-ray.
Method: We retrospectively extracted the medical data of patients with COVID-19 at Taipei Tzu Chi Hospital from May 15th to July 15th in 2021. We recruited patients who received invasive mechanical ventilation. The chest X-ray images of each recruited patient were assigned into four groups (first, before-intubation, post-intubation, and worst). The BRIXIA and percent opacification scores were reviewed by two pulmonologists. To set up a prognostication model, we used the MobilenetV3-Small model with “ImageNet” pretrained weights, followed by high Dropout regularization layers. We practiced the model with Five-Fold Cross-Validation to assess model efficacy.
Result: We finally recruited sixty-four patients. The overall death rate was forty-five percent. The median days since symptom commencement to endotracheal intubation was eight. Age, inferior academic degree, occurrence of COVID-19 complications, and a more severe achievement of the worst chest X-ray were linked to a higher death risk. The accuracy of the first, pre-intubation, post-intubation, and worst chest X-ray by the artificial-intelligence model were 0.88, 0.92, 0.92, and 0.94 respectively. | en_US |