Use of a Novel Clustered Bayesian 3D- CNN Model for Triage Classification of COVID-19 Mass Casualty Incidents
Liege Cheung, Jinrui Cao, Zequn Lin, Anthony C.H. Sek, Adela S.M. Lau · Frontiers in artificial intelligence and applications · 2025
The first author previously conducted an empirical review of the constructs for determining triage recommendations for COVID mass casualty incidents (MCIs). This research aims to continue that previous study and to implement a novel clustered Bayesian 3D-CNN model for predicting COVID-19 MCI triage classification. An experimental research method is used here. Data were collected from the databases of the COVID Tracking Project, Oxford COVID-19 Government Response Tracker, and OxCGRT. The data samples were then integrated to construct a dataset with the three dominants of severity of infection (signs and symptoms of coronavirus), likelihood of spreading (attitudes towards pandemic spreading, personal behaviors, and government policy), and available resources (availability of physicians and medication, hospital vacancies, and evacuation assets) to train a 3D-CNN model on COVID19 test results, recovery histories and medication outcomes. The output of the model includes categories of triage recommendations entitled “immediate (hospital admission with immediate medical care)”, “delayed (quarantine center and observation)”, “minimal (home quarantine with medication)”, “semi-minimal (home quarantine without medication)”, and “no priority (no therapeutic resources)”. Experimental results show that the accuracy rate, AUC, precision, recall and F1-score for the COVID19 MCI triage decisions are 82%, 85.8%, 0.835, 0.789, and 0.811, respectively. In addition, our model outperforms other triage decision models for emergency incidents. As a consequence of this study, medical professionals will be able to apply our model to their hospital data for triage decisions involving COVID-19 or other medical incidents.