Since December 2019, multiple cases of pneumonia due to unknown reasons have emerged in Wuhan, China. Through testing multiple patient samples, scientists extrapolated a new coronavirus termed COVID-19. With no FDA approved therapeutics or treatment available for the disease, diagnosis plays an important role in containing COCVID-19, giving a path to the rapid implementation of control measures to limit the spread. With the disease spreading to almost 100 countries, a million cases have been confirmed worldwide to date. Imaging is one of the main principles used in diagnosing and evaluating the disease, with the final diagnosis depending on reverse transcriptase-polymerase chain reaction (RT-PCR).
In response to the growing number of COVID-19 cases, there is currently a shortage of diagnostic kits worldwide. Multiple industries are coming forward to develop rapid, easy to use diagnostic kits to facilitate testing. However before these kits can be commercialized, they must be tested and validated. With the current available tests taking almost 2 days to complete and produce a result, serial testing is required to rule out any negative cases. Additionally, it is a mystery as to whether an RT-PCR is a gold standard and whether a false positive/ negative result is common. The above reasons highlight the need for alternative testing methods to produce rapid and accurate results to identify, isolate, and treat the affected people.
Chest computed tomography is also a much-used valuable component in testing COVID-19. With some of the patients showing early-stage symptoms in radiological finding, limits the CT ability to differentiate between a positive and negative case. In this current study, the authors have used Artificial Intelligence (AI) algorithms to help in integrating CT scanning in finding the symptoms of the virus, exposure history and reliable lab testing to rapidly diagnose the patients affected with COVID-19.
A trial was performed on 905 patients diagnosed using RT-PCR and next-generation RT-PCR and around 46% (419) people were declared positive for COVID-19. Parallelly in a test set of 279 participants, the AI system managed to achieve accuracy to about 92% of the population and had equal or even better sensitivity than a senior radiologist. The AI system also improved the detection of COVID-19 positive patients with negative CT scans, identifying 17 out of 25 participants who were tested positive via RT-PCR but negative with normal CT scans. In comparison, the radiologists’ declared the said 17 participants to be COVID negative.
AI shows signs of analyzing huge amounts of data quickly, a quality that is much needed in the current pandemic. A major limitation of the above study is the small sample size, with available CT scans and clinical history data, the AI system can help in diagnosing COVID-19 patients rapidly. Though a promising tool, further data collection is required to test the generalization of AI mapping on other patient populations.