AI model diagnoses COVID-19 in routine blood tests with impressive accuracy

AI models can be trained to diagnose COVID-19 using routine laboratory blood tests, according to new research published in Biomedical Signal Processing and Control.

“With new mutations of the virus with higher transmission rates, it is imperative to diagnose positive cases as quickly and accurately as possible,” wrote first author Samin Babaei Rikan, a specialist with the department of computer engineering at Urmia University in Iran, and colleagues. “Therefore, a fast, accurate, and automatic system for COVID-19 diagnosis can be very useful for clinicians.”

The group noted that PCR tests, chest X-rays and chest CT scans are all currently used to diagnose COVID-19—but all three have their own faults that make them less than ideal. To see if they could potentially provide clinicians with another option, Rikan et al. tested the ability of seven machine learning models and four deep learning models to diagnose COVID-19 using routing blood tests. Each model was developed and trained using three different datasets of blood tests that contained positive and negative COVID-19 cases.

Overall, the group had the most consistent success with the deep neural network (DNN) model. For the three datasets, its accuracy was measured as 92.11%, 93.16% and 93.33%, respectively. The model’s sensitivity (96.14%, 93.27% and 77.05%) and specificity (84.56%, 93.02% and 95.27%) for the three different datasets results were also especially strong.