Reconstructing low-dose CT imaging is a challenging mathematics issue known as an ill-posed inverse problem as it involves managing the intrinsic noise in the data. Recent attention has shifted towards deep learning-based techniques in the area of CT image reconstruction. However, these approaches encounter limitations due to extensive data requirements for training and testing. We proposed a novel unsupervised CT reconstruction method that makes use of Attention enhanced Deep image prior (AE-DIP) and show that a neural network with a random initialization can function as a prior. With our approach, the reconstruction procedure is done with the Simultaneous Algebraic Reconstruction Technique (SART), and AE-DIP Deep network is used as regularization prior. During the optimization phase, the model dynamically focus on different sections of the image by integrating attention processes into the DIP framework. This can improve the network’s ability to discriminate between relevant image components and noise, therefore will enhance the quality of the reconstructed images. By combining the benefits of DIP and attention processes, this approach enables more targeted and efficient unsupervised learning throughout the image reconstruction process.