- Artificial Intelligence Algorithm Identifies COVID-19 in CT Scans
- Machine Learning Could Boost Low-Dose CT Scan Performance
The study, published in Nature Communications, showed that this artificial intelligence approach can overcome some of the challenges associated with current COVID-19 testing.
“We demonstrated that a deep learning-based AI approach can serve as a standardized and objective tool to assist healthcare systems as well as patients,” Ulas Bagci, assistant professor in UCF’s department of computer science, said in the study.
“It can be used as a complementary test tool in very specific limited populations, and it can be used rapidly and at large scale in the unfortunate event of a recurrent outbreak.”
Researchers uncovered that the artificial intelligence algorithm could be trained to separate COVID-19 pneumonia in CT scans with 90 percent accuracy.
It can also able to identify positive cases 84 percent of the time and negative cases 93 percent of the time.
Compared to a reverse transcription-polymerase chain reaction, or RT-PCR tests, CT scans can offer deeper insight to not only COVID-19 diagnosis, but also progression of the disease.
“Another benefit to CT scans is that they can detect COVID-19 in people without symptoms, in those who have early symptoms, during the height of the disease and after symptoms resolve,” researchers explained.
As part of the study, researchers trained a computer algorithm to recognize COVID-19 in lung CT scans in 1,280 multinational patients from China, Japan, and Italy.
The algorithm was tested on CT scans of 1,337 patients with lung disease ranging from COVID-19 to cancer and non-COVID pneumonia.
When the computer diagnoses were compared with ones confirmed by physicians, researchers uncovered that the algorithm accurately diagnosed COVID-19-pneumonia in the lungs.
Additionally, the algorithm separated COVID-19 from other diseases, specifically when examining CT scans in the early stages of disease progression.
“We showed that robust AI models can achieve up to 90 percent accuracy in independent test populations, maintain high specificity in non-COVID-19 related pneumonias, and demonstrate sufficient generalizability to unseen patient populations and centers,” Bagci said.
Along with UCF researchers, Baris Turkbey, associate researcher physicians at NIH’s national cancer center institute molecular imaging branch, and Bradford J. Wood, director of The Center for Interventional Oncology and chief of interventional radiology at NIH, participated in the study.
Both NIH and the National Cancer Institute helped to fund the study.
During the COVID-19 pandemic, virtual imaging has been extremely beneficial to both patients and scientists.
Traditionally, COVID-19 lab tests can take up to two days to complete, but CT scans have the potential to analyze large amounts of data quickly.
At the beginning of September, a study published in the American Journal of Roentgenology (AJR), found that the use of virtual imaging trials in assessment and optimization of CT and radiography acquisitions, as well as various analysis tools, can help manage the pandemic.
Researchers first developed a model of patients with COVID-19 and showed how it can be combined with imaging simulators for COVID-19 imaging studies.
Using a specific CT scanner and validated radiography simulator to help illustrate the benefit, researchers virtually imaged three developed COVID-19 computational phantoms, which is a mathematical representation of the human body.
The study found that the simulated abnormalities were realistic regarding shape and texture.
Researchers noted that overall, imaging trials can facilitate the assessment and optimization of imaging methods by imitating the imaging experiment using representative computational models of patients and validated imaging simulators.
By 