Development of Artificial Intelligence (AI) tool to detect COVID-19 condition from a chest X-ray image

Rajalakshmi group of educational institutions in Chennai is familiar for their versatility of academic excellence and research activities. In the group, Centre for Medical Imaging is formed with the objective of creating proficient solutions in medical imaging applications. The centre has members from faculty fraternity, students, and medical imaging professionals. Ever since the onset of the pandemic COVID-19, the members of the centre discusses the research progress across global in the aspect of medical imaging to detect COVID-19 through various imaging modalities like X-Ray. Ultrasound and CT scan. Initially, we were thinking to use ultrasound images but due to some technical issues, we couldn’t follow it up. Later we decided to use chest X-ray images and incidentally, we were able to get around 100 COVID-19 affected subjects’ chest X-ray images through open-source Canadian research forum. With the acquired images we planned to develop an Artificial Intelligence (AI) tool to detect COVID-19 condition from the test chest X-ray image. We decided to add the provision for detecting CAP (Community-Acquired Pneumonia) besides to COVID-19 in the proposed tool. The reason for the inclusion of CAP is that there is a relatively high possibility of getting this condition. We have decided to detect the normal conditions also. Hence, we decided to develop a three-class classifier using that we can detect COVID-19, CAP and normal condition from the test X-ray image. We acquired the necessary CAP & normal chest X-ray images through RSNA (Radiological Society of North America) for training the AI tool. We used 14,148 chest X-ray images of all these three classes for rigorously training the neural network inbuilt in the AI tool. The training process underwent different iterations until we reach a training accuracy of 99.9. In the testing process, 1661 images were used and by adopting different strategies we are able to get a testing accuracy of 95.4. Confusion Matrix Result Parameters: Precision: 97.18% Sensitivity/Recall: 94.51% Specificity: 99.65% Accuracy: 95.4% Deployment of the developed tool in the website is under process. However with the inclusion of more Chest X-ray images of COVID-19 positive cases in the training process, the testing accuracy could be improved. The necessary steps are in the pipeline to acquire more images.
1/11/2021