Detail Increment Module for 3D Reconstruction in Single Particle Cryo-EM

Single particle cryo-EM is currently the mainstream technology for protein 3D reconstruction. In recent years, 3D reconstruction algorithms based on neural networks have developed rapidly. However, when using neural networks, the extremely high noise in the input particle images may result in an incorrect initial model during the initial training phase, leading to a local optimality and affecting the final reconstruction results. To address this issue, we propose an innovative detail increment mechanism module. This module is designed to refine the functioning of the network, enabling it to produce a more accurate initial model during the initial learning phase. Consequently, the reconstruction algorithm can continuously optimize on this basis, ultimately enhancing the resolution of the 3D models reconstructed by the neural network.