Faizan Ahmed, a student in The Lesley H. and Willliam L. Collins College of Professional Studies, along with Fazel Keshtkar, Ph.D., Associate Professor, Division of Computer Science, Mathematics, and Science, and Syed Ahmad Chan Bukhari, Ph.D., Assistant Professor, Division of Computer Science, Mathematics, and Science, and Director, B.S. in Healthcare Informatics, recently published a research article, “A Deep Learning Approach for COVID-19 & Viral Pneumonia Screening with X-ray Images.”
In areas where there is a lack of COVID-19 testing kits, there is an urgent need for alternative diagnostic measures. Given that all hospitals have X-ray machines, it is possible to use X-rays to screen for instances of COVID-19 and viral pneumonia in a patient’s lungs.
In this study, the researchers aimed to identify how effective this deep learning method is in detecting instances of COVID-19 in a patient’s lungs, and if it can be used to distinguish between patients who have viral pneumonia and patients who have normal lung functionality as well.
After performing five-fold cross validation with the proposed deep learning model in this study, an average classification accuracy of 90.64 percent was achieved. The model was able to correctly identify whether a patient has COVID-19, viral pneumonia, or is normal approximately 90 percent of the time.
Using saliency maps, they were able to visualize what areas of the X-ray are of high importance and have the greatest influence on the model’s prediction. These areas can be classified as containing patches of ground-glass opacity, which is critical for early detection of COVID-19 and viral pneumonia. Radiologists can analyze these visualizations to better understand each disease and discover new ways to screen them.
The proposed deep learning model in this study can potentially be used as an alternative, or in conjunction with, the standard RT-PCR testing methods currently used to increase efficiency and efficacy in detecting positive cases of both COVID-19 and viral pneumonia. Given that RT-PCR testing methods are time-consuming, laborious, and can have high rates of false negative results, this model offers a fast, automated, end-to-end solution for testing.