Automated grading of chest x-ray images for viral pneumonia with convolutional neural networks ensemble and region of interest localization

• DOI: 10.1371/journal.pone.0280352
• Publication Date: 2023-01-18T04:00:00Z
• Authors: Asad Khan; Muhammad Usman Akram; Sajid Nazir
• Keywords: Grading (engineering); Transfer of learning

Following its initial identification on December 31, 2019, COVID-19 quickly spread around the world as a pandemic claiming more than six million lives. An early diagnosis with appropriate intervention can help prevent deaths and serious illness distinguishing symptoms that set apart from pneumonia influenza frequently don't show up until after patient has already suffered significant damage. A chest X-ray (CXR), one of many imaging modalities are useful for detection most used, offers non-invasive method detection. The CXR image analysis also reveal additional disorders, such pneumonia, which anomalies in lungs. Thus these CXRs be used automated grading aiding doctors making better diagnosis. In order to classify into Negative Pneumonia, Typical, Indeterminate, Atypical, we publicly available competition dataset SIIM-FISABIO-RSNA Kaggle. suggested architecture employed an ensemble EfficientNetv2-L classification, was trained via transfer learning initialised weights ImageNet21K various subsets data (Code proposed methodology is at: https://github.com/asadkhan1221/siim-covid19.git). To identify localise opacities, YOLO combined using Weighted Boxes Fusion (WBF). Significant generalisability gains were made possible by technique's addition classification auxiliary heads CNN backbone. improved further utilising test time augmentation both classifiers localizers. results Mean Average Precision score deep model achieves 0.617 0.609 public private sets respectively comparable other techniques Kaggle dataset.