Lung Region Segmentation Using Modified U-Net Architecture

Abstract

In order to combat the continuing COVID-19 pandemic, early infection diagnosis is essential. Lung infections can be detected using a screening technique called chest X-ray imaging (CXR). During pandemics, integrating machine learning techniques with medicine analysis is crucial in relieving the enormous load on healthcare systems and clinicians. As the continuing COVID-19 crisis intensifies in countries with dense populations and few testing kits, radiological Imaging can serve as a crucial diagnostic tool to properly classify covid-19 patients and administer the appropriate medication on time. In light of this objective, we describe our Research on segmenting lung areas utilizing a deep learning architecture and chest X-ray images as the source material. We used 2 public datasets, Montgomery & Shenzhen and Covid-Qu dataset. Also, the Research created the KURD-covid dataset and collected 1300 x-ray images from local health care centres that contain two labels, covid-19 and normal x-ray with manual masks for all images. We proposed the segmentation model based on state-of-art U-Net architecture. The result of the Covid-Qu daset was., %97 IoU and %99.5 DSC, and in KURD-covid is %71 IoU and %82 DSC.