COVID-19 Lungs CT Scan Lesion Segmentation

  • Muhammad Shariq Shoaib University of the Punjab, Lahore, Pakistan
  • Zobia Suhail University of the Punjab, Lahore, Pakistan
Keywords: Pandemic, COVID-19, CT Scan, Image Processing, Lesion, Mediastinal Window, U-Net, Segmentation


The outburst of the novel coronavirus 2019 has caused a multinational pandemic that has impacted a huge number of individuals around the globe. One of the primary indications of COVID-19 is the formation of lesions in the lungs, which can cause severe harm to the respiratory system and lead to death. In the following study, we submitted a novel strategy for making lung window CT scans and mediastinal window CT scans similar, to input it into a customized U-Net based model to achieve a decent degree of accuracy in segmenting these lung lesions. The method suggested in this research study is based on specialized image processing algorithms to normalize the CT scans' pixel intensity level and uniform the mediastinal and lung window CT scans. This allows us to accurately segment the lung lesions using a UNet model with a single channel input. We were able to achieve an IOU score of 82.4%, which is a significant addition to the existing Medical World. Additionally, the suggested approach is on par with cutting-edge methods.