3D Volume Reconstruction from Sparse 2D CT Slices Using Lightweight CNN & Attention

Abstract

Sparse slice acquisition is used to a great extent in the low dose CT imaging. This assistsin reduction of radiation. Nonetheless, it is likely to damage the quality of 3D reconstruction significantly. The resultant artifacts and noise are manifested. There is some evidence of a lightweight 2D CNN auto encoder, which consists of channel attention. The purpose of this set up is to enhance the quality of the sparse CT slices. It reinstates important structural information. Noise gets reduced too. All this is prior to the slices entering the entire volumetric process. Linear interpolation is used after the reconstruction process. This forms a full volume in 3D. The continuity across the slices is enhanced in the process. Experimental reviews point to high performance in this. The average SSIM of the model is 0.9660. Average PSNR comes in at 25.99 dB. These figures point to good structural fidelity in the general. Distortion is also rather low. Such outcomes are indicative of the usefulness of the lightweight approach. It is effective in efficient sparse data 3D CT reconstruction.