Kapitel/Beitrag aus dem Buch: Schulte, H et al. 2020. Proceedings – 30. Workshop Computational Intelligence : Berlin, 26. – 27. November 2020.
SARS-CoV-2 is a highly contagious virus that can induce pulmonary complications like viral pneumonia and acute respiratory distress syndrome (ARDS). In order to support RT-PCR testing, chest X-rays are used to identify the presence of COVID-19 in lungs. Radiologists proficient in chest X-Ray (CXR) interpretation are scarce, motivating our work of examining the performanceof convolutional neural networks (CNNs) on this task. CNNs are the state-ofthe- art image classification method. In this work we classify X-rays into four classes (COVID-19, other lung opacity, other diseases and normal). We find, that MobileNetV1 is the best CNN for this task and achieve an overall accuracy of 70% and a COVID-19 accuracy of 83% on a test data set. By increasing the number of images of the unrestricted classes normal, lung opacity and other, the COVID-19 accuracy can be increased to 95% and the overall accuracy to 78%. We leverage model interpretability techniques and provide attention heatmaps that can assist in validating the model’s decision process.
Kurz, P et al. 2020. On the Detection of SARS-CoV-2 induced Pneumonia in X-Ray Thorax Images with Convolutional Neural Networks. In: Schulte, H et al (eds.), Proceedings – 30. Workshop Computational Intelligence : Berlin, 26. – 27. November 2020. Karlsruhe: KIT Scientific Publishing. DOI: https://doi.org/10.58895/ksp/1000124139-7
This chapter distributed under the terms of the Creative Commons Attribution + ShareAlike 4.0 license. Copyright is retained by the author(s)
Dieses Buch ist Peer reviewed. Informationen dazu Hier finden Sie mehr Informationen zur wissenschaftlichen Qualitätssicherung der MAP-Publikationen.
Veröffentlicht am 20. November 2020