الفهرس | Only 14 pages are availabe for public view |
Abstract As COVID 19 spreads all over the world, researchers use all the available resources to diagnose it as fast as possible. Lung CT images are one of the most effective ways to diagnosis COVID 19. As CT scanning of lung provides a 3D representation of lung and has a high sensitivity.Also, CT is fast and low-cost method compared to RT-PCR kits which makes it an effective method to diagnosis COVID 19.An automatic segmentation model needed to make diagnosis easier and take precautions measures as earlier as possible to reduce the spreading. The main aim of thesis was to design a novel, reliable, accurate and automatic semantic segmentation network images to segment COVID-19 lesion with the help of chest CT images.In this thesis, two proposed system were implemented.Model 1 was DeepLabV3+ integrated with MobileNet-V2 which combine between the DeepLabV3+ semantic segmentation and the MobileNet-V2 which used mostly in classification problems. MobileNet-V2 used in Model 1 as an encoder of the segmentation network, it used inverted residuals and linear bottlenecks as its backbone which based on using depth-wise separable convolution layers rather than standard convolution. Therefore, the use of MobileNet-V2 solved the Gradients’ vanishing and exploding problem as it has a residual connection between the thin bottlenecks of each block and saved the number multiplication operation due to the use of depth-wise separable convolution layer. |