Accepted for/Published in: JMIR Formative Research
Date Submitted: Feb 27, 2023
Date Accepted: Dec 29, 2023
Artificial Intelligence Scoring Performance Compared to Brixia Score on Chest X-Ray Against Disease Severity in Suspected Coronavirus Disease-2019 Patients: A diagnostic accuracy study
ABSTRACT
Background:
Background:
Artificial intelligence (AI) has been known to increase the precision of COVID-19 pneumonia diagnosis using chest X-rays (CXRs).
Objective:
Objective:
This study aimed to determine and compare the performance of AI scoring system using a color heat map vs. the Brixia scoring system (overall CXR score) on CXR to screen and predict the severity of COVID-19 pneumonia.
Methods:
Methods:
This was a cross-sectional study, involving 203 suspected and 97 RT-PCR-confirmed COVID-19 patients. Chest X-rays were quantitatively assessed using the CAD4COVID software and semi-quantitatively using the Brixia scoring system. Performance analyses were assessed using the area under the curve (AUC) estimation and comparison between AUCs, as well as comparisons of sensitivity, specificity, positive predictive value, negative predictive value, and accuracy.
Results:
Results:
AI probability score, AI affected lung area (ALA) score, and overall CXR score could not discriminate RT-PCR results of suspected COVID-19 patients. AI probability score (AUC = 0.888; 95% CI 0.820-0.956), AI ALA score (AUC = 0.875, 95% IK 0.789-0.953) and overall CXR score (AUC = 0.878, 95% CI 0.808-0.948) had excellent agreement to determine disease severity in subjects with confirmed COVID-19. AI probability score (Sn 87.2%, Acc 85.6%) and AI ALA score (Sn 82.6%, Acc 80.4%) were more sensitive and accurate than the Brixia CXR score (Sn 75.6%, Acc 78.4%) to determine the severity of COVID-19 pneumonia.
Conclusions:
Conclusions:
AI and Brixia scoring systems had fair to excellent discriminatory ability for disease severity. Both also had excellent performance in predicting disease severity, with better prediction seen in AI, albeit non – significant
Citation
Request queued. Please wait while the file is being generated. It may take some time.