Fillnet: A cryogenic electron tomography restoration framework integrating FFT_Unet architecture and weight...

Cryogenic electron tomography is an important technique that enables the three-dimensional visualisation of microscopic samples. In cryogenic electron tomography, a series of two-dimensional projection images is acquired from different tilt angles of the sample and computationally reconstructed into a tomogram. The tilt range of the specimen stage is typically limited to a certain angular range. Beyond this range, the sample may become too thick for electrons to penetrate, and mechanical components such as the support grid or holder may obstruct the beam, resulting in a loss of image quality. This angular limitation leads to missing information in the reconstructed tomograms, known as the missing wedge problem. Moreover, the use of low-dose electron imaging and other experimental constraints introduces considerable noise, thereby reducing the signal-to-noise ratio of the reconstructed tomogram. In order to solve the problems of missing wedges and low signal-to-noise ratio of tomograms, the Fillnet tomogram restoration framework was designed in this study. The training pair generation module and the FFT_Unet model are specially designed in this framework to improve the accurate acquisition of three-dimensional features in tomograms. Different loss functions are also designed to improve the model's attention to the special features of the samples.