U-Net Model Combined with Ensemble Learning for Segmentation of Covid-19 Computed Tomography Images

Covid-19 CT lung segmentation has been an important area of study for researchers since early 2020, when the Covid-19 pandemic became widespread. In the early days of the pandemic, few studies were conducted due to problems arising both in the diagnosis phase of the disease and in accessing patient data. Since the sharing of Covid-19 data obtained from different imaging modalities, studies in this field have also increased. In this study, segmentation of infected areas in Covid-19 Computed Tomography lung images is performed. After the data is first processed with data augmentation techniques, it is transferred to the U-Net model, where pretrained models are used on the encoder side. The combination process is performed by ensemble learning among the models used by the encoder. The results obtained with the proposed method were found to be more successful in terms of Covid-19 segmentation compared to other studies.