Deep Learning Techniques for the Radiological Imaging of COVID-19 PDF Download

Author: Robert Hertel
Publisher:
ISBN:
Category :
Languages : en
Pages : 0

Get Book Here

Book Description
The AI research community has recently been intensely focused on diagnosing COVID19 by applying deep learning technology to the X-ray scans taken of COVID-19 patients. COVID-19 shares many of the same imaging characteristics as other common forms of bacterial and viral pneumonia. Differentiating COVID-19 from other common pulmonary infections, therefore, is a non-trivial task. While RT-PCR tests are the first viral tests commonly performed on COVID-19 patients, radiological tests are often reserved for further study of the illness in patients presenting with increased risk factors. To help offset what commonly requires hours of tedious manual annotation, our work uses Convolutional Neural Networks and other machine learning techniques to decrease the time radiologists spend interpreting COVID-19 radiological scans. Deep learning experts commonly use transfer learning to offset the small number of images typically available in medical imaging tasks. Our first study's architecture included a deep neural network that was pretrained on over one hundred thousand X-ray images. We incorporated this architecture into two models with the purpose of diagnosing COVID-19. The experimental results demonstrate the robustness of our deep learning models, ultimately achieving sensitivities of 95% and 96% for our three-class and two-class models respectively. To help further clarify the diagnosis of suspected COVID-19 patients, in our second study, we have designed a deep learning pipeline with a segmentation module and ensemble classifier. After performing a thorough comparative analysis, we demonstrate that our best model can successfully obtain an accuracy of 91% and a sensitivity of 92%. Following a detailed description of our deep learning pipeline, we present the strengths and shortcomings of our approach and compare our model with other similarly constructed models. Finally, we conclude with possible future directions for this research.