Fast Three-Dimensional Noise Reduction for Real-Time Electron Tomography

Abstract

Electron tomography (ET) is the leading imaging technique for visualizing the molecular architecture of complex biological specimens. Real-time ET systems allow scientists to acquire experimental datasets and obtain a preliminary version of three-dimensional structure of the specimen. This rough structure allows assessment of the quality of the sample and can also be used as a guide to collect more datasets. However, the low signal-to-noise ratio of the ET datasets precludes detailed interpretation and makes their assessment difficult. Therefore, noise reduction methods should be integrated in these real-time ET systems for their full exploitation. However, feature-preserving noise reduction methods are typically computationally intensive, which hinders real-time response. This work proposes and evaluates fast implementations of a sophisticated noise reduction method with capabilities of preservation of biologically relevant features. These implementations are designed to exploit the high performance computing (HPC) capabilities of modern multicore platforms and of graphics processing units. It is shown that the use of HPC on modern platforms makes this noise reduction method able to provide datasets appropriate for assessment in a matter of seconds, thereby making it suitable for integration in current real-time ET systems.