GPU Accelerated 3D Tomographic Reconstruction and Visualization from Noisy Electron Microscopy Tilt-Series (original) (raw)
Related papers
A distributed multi-GPU system for high speed electron microscopic tomographic reconstruction
Ultramicroscopy, 2011
Full resolution electron microscopic tomographic (EMT) reconstruction of large-scale tilt series requires significant computing power. The desire to perform multiple cycles of iterative reconstruction and realignment dramatically increases the pressing need to improve reconstruction performance. This has motivated us to develop a distributed multi-GPU (graphics processing unit) system to provide the required computing power for rapid constrained, iterative reconstructions of very large three-dimensional (3D) volumes. The participating GPUs reconstruct segments of the volume in parallel, and subsequently, the segments are assembled to form the complete 3D volume. Owing to its power and versatility, the CUDA (NVIDIA, USA) platform was selected for GPU implementation of the EMT reconstruction. For a system containing 10 GPUs provided by 5 GTX295 cards, 10 cycles of SIRT reconstruction for a tomogram of 4096(2) × 512 voxels from an input tilt series containing 122 projection images of 4...
Iterative reconstruction algorithms are becoming increasingly important in electron tomography of biological samples. These algorithms, however, impose major computational demands. Parallelization of the computations must be employed to maintain acceptable running times. Graphics Processing Units (GPUs) have been demonstrated to be highly cost-effective for carrying out these computations with a high degree of parallelism. In a recent paper by Xu et al.
Journal of Structural Biology, 2010
Iterative reconstruction algorithms pose tremendous computational challenges for 3D Electron Tomography (ET). Similar to X-ray Computed Tomography (CT), graphics processing units (GPUs) offer an affordable platform to meet these demands. In this paper, we outline a CT reconstruction approach for ET that is optimized for the special demands and application setting of ET. It exploits the fact that ET is typically cast as a parallel-beam configuration, which allows the design of an efficient data management scheme, using a holistic sinogram-based representation. Our method produces speedups of about an order of magnitude over a previously proposed GPU-based ET implementation, on similar hardware, and completes an iterative 3D reconstruction of practical problem size within minutes. We also describe a novel GPU-amenable approach that effectively compensates for reconstruction errors resulting from the TEM data acquisition on (long) samples which extend the width of the parallel TEM beam. We show that the vignetting artifacts typically arising at the periphery of non-compensated ET reconstructions are completely eliminated when our method is employed.
Local regularization of tilt projections reduces artifacts in electron tomography
Electron tomography produces very high resolution 3D image volumes useful for investigating the structure and function of cellular components. Unfortunately, unavoidable discontinuities and physical constraints in the acquisition geometry lead to a range of artifacts that can affect the reconstructed image. In particular, highly electron dense regions, such as gold nanoparticles, can hide proximal biological structures and degrade the overall quality of the reconstructed tomograms. In this work we introduce a pre-reconstruction non-conservative non-linear isotropic diffusion (NID) filter that automatically identifies and reduces local irregularities in the tilt projections. We illustrate the improvement in quality obtained using this approach for reconstructed tomograms generated from samples of malaria parasite-infected red blood cells. A quantitative and qualitative evaluation for our approach on both simulated and real data is provided.
A fast reconstruction algorithm for electron microscope tomography
Journal of Structural Biology, 2003
We have implemented a Fast Fourier Summation algorithm for tomographic reconstruction of three-dimensional biological data sets obtained via transmission electron microscopy. We designed the fast algorithm to reproduce results obtained by the direct summation algorithm (also known as filtered or R-weighted backprojection). For two-dimensional images, the new algorithm scales as OðN h M log MÞ þ OðMN log N Þ operations, where N h is the number of projection angles and M Â N is the size of the reconstructed image. Three-dimensional reconstructions are constructed from sequences of two-dimensional reconstructions. We demonstrate the algorithm on real data sets. For typical sizes of data sets, the new algorithm is 1.5-2.5 times faster than using direct summation in the space domain. The speed advantage is even greater as the size of the data sets grows. The new algorithm allows us to use higher order spline interpolation of the data without additional computational cost. The algorithm has been incorporated into a commonly used package for tomographic reconstruction.
High-Performance Blob-Based Iterative Reconstruction of Electron Tomography on Multi-GPUs
Three-dimensional (3D) reconstruction of electron tomography (ET) has emerged as a leading technique to elucidate the molecular structures of complex biological specimens. Blob-based iterative methods are advantageous reconstruction methods for D reconstruction of ET, but demand huge computational costs. Multiple Graphic processing units (multi-GPUs) offer an affordable platform to meet these demands, nevertheless, are not efficiently used owing to a synchronous communication scheme and the limited available memory of GPUs. We propose a multilevel parallel scheme combined with an asynchronous communication scheme and a blob-ELLR data structure. The asynchronous communication scheme is used to minimize the idle GPU time. The blob-ELLR data structure only needs nearly 1/16 of the storage space in comparison with ELLPACK-R (ELLR) data structure and yields significant acceleration. Experimental results indicate that the multilevel parallel scheme allows efficient implementations of D re...
Wavelet Frame Based Algorithm for 3D Reconstruction in Electron Microscopy
SIAM Journal on Scientific Computing, 2014
In electron microscopy, three-dimensional (3D) reconstruction is one key component in many computerized techniques for solving 3D structures of large protein assemblies using electron microscopy images of particles. Main challenges in 3D reconstruction include very low signal-to-noise ratio and very large scale of data sets involved in the computation. Motivated by the recent advances of sparsity-based regularization in the wavelet frame domain for solving various linear inverse problems in imaging science, we proposed a wavelet tight frame based 3D reconstruction approach that exploits the sparsity of the 3D density map in a wavelet tight frame system. The proposed approach not only runs efficiently in terms of CPU time but also requires a much lower memory footprint than existing framelet-based regularization methods. The convergence of the proposed iterative scheme and the functional it minimizes is also examined, together with the connection to existing wavelet frame based regularizations. The numerical experiments showed good performance of the proposed method when it is used in two electron microscopy techniques: the single particle method and electron tomography.
1997 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2000
In electron tomographic reconstructions of biological specimens the information about their structure is not directly accessible since most of the signal is buried in noise. An interpretation of the images using surface and volume rendering techniques is difficult due to the noise sensitivity of rendering algorithms. We explore the use of various multiscale representations for denoising 2D and 3D images. Orthogonal wavelet transforms applied to multidimensional data exhibit poor results due to the lack of translational and directional invariance. Extending the 1D translation-invariant denoising algorithm of Coifman and Donoho to higher dimensions proves to overcome the poor performance of orthogonal wavelet transforms. We present a method to quantify the loss of information due to denoising artifacts on data with an unknown signalnoise relationship, and propose a scheme for denoising of such data. Experiments show invariant wavelet denoising to perform well in reconstructing signals out of noisy 3D data while preserving most of the actual information.
An interactive ImageJ plugin for semi-automated image denoising in electron microscopy
Nature Communications, 2020
The recent advent of 3D in electron microscopy (EM) has allowed for detection of nanometer resolution structures. This has caused an explosion in dataset size, necessitating the development of automated workflows. Moreover, large 3D EM datasets typically require hours to days to be acquired and accelerated imaging typically results in noisy data. Advanced denoising techniques can alleviate this, but tend to be less accessible to the community due to low-level programming environments, complex parameter tuning or a computational bottleneck. We present DenoisEM: an interactive and GPU accelerated denoising plugin for ImageJ that ensures fast parameter tuning and processing through parallel computing. Experimental results show that DenoisEM is one order of magnitude faster than related software and can accelerate data acquisition by a factor of 4 without significantly affecting data quality. Lastly, we show that image denoising benefits visualization and (semi-)automated segmentation a...
Real-Time Electron Tomography Based on GPU Computing
Lecture Notes in Computer Science, 2011
Electron tomography (ET) has emerged as the leading technique for the structural analysis of unique complex biological specimens. Recently, real-time ET systems have appeared on the scene and they combine the computer-assisted image collection with the 3D reconstruction, and provide the users a preliminary structure of the specimen. This rough structure allows the users to easily evaluate the quality of the specimen and decide whether a more time-consuming processing and thorough analysis of the dataset is worthwhile. The aim of this work is to develop software for real-time ET systems. The principle of ET is based upon 3D reconstruction from projections. By means of tomographic reconstruction algorithms, the projection images in the tilt series can then be combined to yield the 3D structure of the specimen.The 3D structure has poor signal to noise ratio, so it is necessary an additional non linear filtering process in order to achieve enough resolution. Then, Matrix Weighted Back Projections (Matrix WBP) and Beltrami methods have been selected as reconstruction and filter procedures, respectively. First the Matrix WBP is applied to the input sinograms to obtain the three-dimensional structure and, next, Beltrami filter de-noises the image. Both methods are highly accelerated by GPU platforms. The power of GPU computing is then exploited to further improve the performance and yield reconstructions of biological datasets in seconds, it allows to integrate both methods on real time electron tomography systems.