Comparitive Analysis of CT Image Reconstruction using Deep Learning Layer Structures

CT scanning yields useful information on the internal anatomy or pathology of the human body. Researchers are always working to use low radiation doses to create CT that is both high-quality and efficient. Originally, the CT image was rebuilt from raw projection data using FBP (filtered back projection) and IR (iterative reconstruction). However, deep learning techniques change the whole architectures manifold. CT reconstruction became more intriguing to study in order to produce high-quality images while utilizing low-dose projection data and reducing computational costs. Raw projection data (2016 NIH-AAPM-Mayo Clinic) for reconstruction is available online. In this work a detailed comparative analysis has been carried out between different deep learning technique and an approach has been proposed for reconstructing a high quality image. In proposed methodology deep learning technique will first be used to rebuild the sinogram from slices, and after that, the entire image will be created. Remarkably, efficient deep learning algorithms produce images at the expense of computing efficiency. After the reconstruction process second step would be preprocessing using bilateral filter to further minimize the noise and preserve the edges while last step is the segmentation phase in which different body parts in the CT scan would be segmented to visualize the scan more deeply and accurately.