The standard practices for screening patients with COVID-19 are the CT imaging or RT-PCR (real-time polymerase chain reaction) for testing viral nucleic acid, which is expensive and in need of professional equipment and waiting time.
Fudan University in Shanghai, China has suggested a method that have shown that the COVID-19 patients are usually accompanied by ocular manifestations and constructed a machine-learning model based on ocular surface features and proposed a new screening method for COVID-19.
A retrospective study of analyzing 446 subjects and a prospective study with 128 subjects were conducted with this method.
The performance was measured at receiver-operating-characteristic curve (AUC), sensitivity, specificity and accuracy. Results The performance of detecting COVID-19 patients in the retrospective study have achieved an AUC of 0.999 (95% CI, 0.997-1.000), with a sensitivity of 0.982 (95% CI, 0.954-1.000), and a specificity of 0.978 (95% CI, 0.961-0.995). And in the prospective study, our model performance on COVID-19 has achieved an AUC of 0.980 (95% CI, 0.970-0.990), with a sensitivity of 0.770 (95% CI, 0.694-0.846), and a specificity of 0.973(95% CI, 0.957-0.989).
Conclusion – This deep learning method based on eye-region images demonstrates the high accuracy to distinguish COVID-19 patients.
Yanwei Fu, a computer scientist at Fudan University in Shanghai, China suggested this method and has published article about it.