CovSegNet: A Multi Encoder–Decoder Architecture for Improved Lesion Segmentation of COVID-19 Chest CT Scans

Automatic lung lesion segmentation of chest computer tomography (CT) scans is considered a pivotal stage toward accurate diagnosis and severity measurement COVID-19. Traditional U-shaped encoder–decoder architecture its variants suffer from diminutions contextual information in pooling/upsampling operations with increased semantic gaps among encoded decoded feature maps as well instigate vanishing gradient problems for sequential propagation that result suboptimal performance. Moreover, operating 3-D CT volume poses further limitations due to the exponential increase computational complexity making optimization difficult. In this article, an automated COVID-19 scheme proposed utilizing highly efficient neural network architecture, namely CovSegNet, overcome these limitations. Additionally, two-phase training introduced where deeper 2-D employed generating region-of-interest (ROI)-enhanced followed by shallower enhancement more without increasing burden. Along traditional vertical expansion Unet, we have horizontal multistage modules achieving optimum multiscale are integrated into scale transition process loss information. fusion module pyramid reduce between subsequent encoder/decoder while facilitating parallel propagation. Outstanding performances been achieved three publicly available datasets largely outperform other state-of-the-art approaches. The can be easily extended wide variety applications. Impact Statement—With lower sensitivity (60–70%), elongated testing time, dire shortage kits, RTPCR based diagnostic heavily relies on postCT manual inspection investigation. Hence, automating infected lesions extraction chestCT volumes will major progress faster However, challenging conditions diffused, blurred, varying shaped edges lesions, conventional approaches fail provide precise deleterious false estimation incorporating (CovSegNet) overcomes significant improvement performance (8.4% averaged dice scale) over two datasets. Therefore, effective, economical tool physicians infection analysis greatly spread massive death toll deadly virus through mass-screening.