Accelerating iterative CT reconstruction algorithms using Tensor Cores

Journal of Real-Time Image Processing ◽

10.1007/s11554-020-01069-5 ◽

2021 ◽

Author(s):

Mohsen Nourazar ◽

Bart Goossens

Keyword(s):

Machine Learning ◽

Real Time ◽

Programming Model ◽

Matrix Multiplication ◽

Hardware Acceleration ◽

Parallel Projection ◽

Ct Reconstruction ◽

Reconstruction Algorithms ◽

Reconstruction Methods ◽

Mixed Precision

AbstractTensor Cores are specialized hardware units added to recent NVIDIA GPUs to speed up matrix multiplication-related tasks, such as convolutions and densely connected layers in neural networks. Due to their specific hardware implementation and programming model, Tensor Cores cannot be straightforwardly applied to other applications outside machine learning. In this paper, we demonstrate the feasibility of using NVIDIA Tensor Cores for the acceleration of a non-machine learning application: iterative Computed Tomography (CT) reconstruction. For large CT images and real-time CT scanning, the reconstruction time for many existing iterative reconstruction methods is relatively high, ranging from seconds to minutes, depending on the size of the image. Therefore, CT reconstruction is an application area that could potentially benefit from Tensor Core hardware acceleration. We first studied the reconstruction algorithm’s performance as a function of the hardware related parameters and proposed an approach to accelerate reconstruction on Tensor Cores. The results show that the proposed method provides about 5 $$\times $$ × increase in speed and energy saving using the NVIDIA RTX 2080 Ti GPU for the parallel projection of 32 images of size $$512\times 512$$ 512 × 512 . The relative reconstruction error due to the mixed-precision computations was almost equal to the error of single-precision (32-bit) floating-point computations. We then presented an approach for real-time and memory-limited applications by exploiting the symmetry of the system (i.e., the acquisition geometry). As the proposed approach is based on the conjugate gradient method, it can be generalized to extend its application to many research and industrial fields.

Download Full-text

A Benchmark Dataset and Deep Learning-Based Image Reconstruction for Electrical Capacitance Tomography

Sensors ◽

10.3390/s18113701 ◽

2018 ◽

Vol 18 (11) ◽

pp. 3701 ◽

Cited By ~ 9

Author(s):

Jin Zheng ◽

Jinku Li ◽

Yi Li ◽

Lihui Peng

Keyword(s):

Machine Learning ◽

Image Reconstruction ◽

Benchmark Dataset ◽

Electrical Capacitance Tomography ◽

Electrical Capacitance ◽

Reconstruction Algorithms ◽

Reconstruction Methods ◽

Image Reconstruction Algorithms ◽

Capacitance Tomography ◽

The Relationship

Electrical Capacitance Tomography (ECT) image reconstruction has developed for decades and made great achievements, but there is still a need to find a new theoretical framework to make it better and faster. In recent years, machine learning theory has been introduced in the ECT area to solve the image reconstruction problem. However, there is still no public benchmark dataset in the ECT field for the training and testing of machine learning-based image reconstruction algorithms. On the other hand, a public benchmark dataset can provide a standard framework to evaluate and compare the results of different image reconstruction methods. In this paper, a benchmark dataset for ECT image reconstruction is presented. Like the great contribution of ImageNet that transformed machine learning research, this benchmark dataset is hoped to be helpful for society to investigate new image reconstruction algorithms since the relationship between permittivity distribution and capacitance can be better mapped. In addition, different machine learning-based image reconstruction algorithms can be trained and tested by the unified dataset, and the results can be evaluated and compared under the same standard, thus, making the ECT image reconstruction study more open and causing a breakthrough.

Download Full-text

Low Complexity FFT Factorization for CS Reconstruction

International Journal of Engineering and Advanced Technology - Regular Issue ◽

10.35940/ijeat.c4675.029320 ◽

2020 ◽

Vol 9 (3) ◽

pp. 438-442

Keyword(s):

Real Time ◽

High Speed ◽

Performance Metrics ◽

Matching Pursuit ◽

Matrix Multiplication ◽

Low Complexity ◽

Orthogonal Matching Pursuit ◽

Reconstruction Algorithms ◽

Wide Range ◽

Computationally Intensive

In this paper is presented a novel area efficient Fast Fourier transform (FFT) for real-time compressive sensing (CS) reconstruction. Among various methodologies used for CS reconstruction algorithms, Greedy-based orthogonal matching pursuit (OMP) approach provides better solution in terms of accurate implementation with complex computations overhead. Several computationally intensive arithmetic operations like complex matrix multiplication are required to formulate correlative vectors making this algorithm highly complex and power consuming hardware implementation. Computational complexity becomes very important especially in complex FFT models to meet different operational standards and system requirements. In general, for real time applications, FFT transforms are required for high speed computations as well as with least possible complexity overhead in order to support wide range of applications. This paper presents an hardware efficient FFT computation technique with twiddle factor normalization for correlation optimization in orthogonal matching pursuit (OMP). Experimental results are provided to validate the performance metrics of the proposed normalization techniques against complexity and energy related issues. The proposed method is verified by FPGA synthesizer, and validated with appropriate currently available comparative analyzes.

Download Full-text

Degradation-Aware Deep Learning Framework for Sparse-View CT Reconstruction

Tomography ◽

10.3390/tomography7040077 ◽

2021 ◽

Vol 7 (4) ◽

pp. 932-949

Author(s):

Chang Sun ◽

Yitong Liu ◽

Hongwen Yang

Keyword(s):

Deep Learning ◽

Structural Similarity ◽

Frequency Component ◽

Ct Images ◽

Training Dataset ◽

Ct Reconstruction ◽

Reconstruction Algorithms ◽

Image Domain ◽

Learning Framework ◽

Reconstruction Methods

Sparse-view CT reconstruction is a fundamental task in computed tomography to overcome undesired artifacts and recover the details of textual structure in degraded CT images. Recently, many deep learning-based networks have achieved desirable performances compared to iterative reconstruction algorithms. However, the performance of these methods may severely deteriorate when the degradation strength of the test image is not consistent with that of the training dataset. In addition, these methods do not pay enough attention to the characteristics of different degradation levels, so solely extending the training dataset with multiple degraded images is also not effective. Although training plentiful models in terms of each degradation level can mitigate this problem, extensive parameter storage is involved. Accordingly, in this paper, we focused on sparse-view CT reconstruction for multiple degradation levels. We propose a single degradation-aware deep learning framework to predict clear CT images by understanding the disparity of degradation in both the frequency domain and image domain. The dual-domain procedure can perform particular operations at different degradation levels in frequency component recovery and spatial details reconstruction. The peak signal-to-noise ratio (PSNR), structural similarity (SSIM) and visual results demonstrate that our method outperformed the classical deep learning-based reconstruction methods in terms of effectiveness and scalability.

Download Full-text

Efficient Hyper-Parameter Selection in Total Variation-Penalised XCT Reconstruction Using Freund and Shapire’s Hedge Approach

Mathematics ◽

10.3390/math8040493 ◽

2020 ◽

Vol 8 (4) ◽

pp. 493 ◽

Cited By ~ 1

Author(s):

Stéphane Chrétien ◽

Manasavee Lohvithee ◽

Wenjuan Sun ◽

Manuchehr Soleimani

Keyword(s):

Machine Learning ◽

Total Variation ◽

Cross Validation ◽

Computational Effort ◽

Parameter Selection ◽

Reconstruction Algorithms ◽

Reconstruction Methods ◽

Learning Procedure ◽

Selection Approach ◽

Type Machine

This paper studies the problem of efficiently tuning the hyper-parameters in penalised least-squares reconstruction for XCT. Discovered through the lens of the Compressed Sensing paradigm, penalisation functionals such as Total Variation types of norms, form an essential tool for enforcing structure in inverse problems, a key feature in the case where the number of projections is small as compared to the size of the object to recover. In this paper, we propose a novel hyper-parameter selection approach for total variation (TV)-based reconstruction algorithms, based on a boosting type machine learning procedure initially proposed by Freund and Shapire and called Hedge. The proposed approach is able to select a set of hyper-parameters producing better reconstruction than the traditional Cross-Validation approach, with reduced computational effort. Traditional reconstruction methods based on penalisation can be made more efficient using boosting type methods from machine learning.

Download Full-text

FPGA Implementation of Real-Time Compressive Sensing with Partial Fourier Dictionary

International Journal of Antennas and Propagation ◽

10.1155/2016/1671687 ◽

2016 ◽

Vol 2016 ◽

pp. 1-12 ◽

Cited By ~ 11

Author(s):

Yinghui Quan ◽

Yachao Li ◽

Xiaoxiao Gao ◽

Mengdao Xing

Keyword(s):

Compressive Sensing ◽

Real Time ◽

Graphics Processing Units ◽

Large Scale ◽

Matching Pursuit ◽

Matrix Decomposition ◽

Hardware Acceleration ◽

Least Square ◽

Reconstruction Algorithms ◽

Least Square Problem

This paper presents a novel real-time compressive sensing (CS) reconstruction which employs high density field-programmable gate array (FPGA) for hardware acceleration. Traditionally, CS can be implemented using a high-level computer language in a personal computer (PC) or multicore platforms, such as graphics processing units (GPUs) and Digital Signal Processors (DSPs). However, reconstruction algorithms are computing demanding and software implementation of these algorithms is extremely slow and power consuming. In this paper, the orthogonal matching pursuit (OMP) algorithm is refined to solve the sparse decomposition optimization for partial Fourier dictionary, which is always adopted in radar imaging and detection application. OMP reconstruction can be divided into two main stages: optimization which finds the closely correlated vectors and least square problem. For large scale dictionary, the implementation of correlation is time consuming since it often requires a large number of matrix multiplications. Also solving the least square problem always needs a scalable matrix decomposition operation. To solve these problems efficiently, the correlation optimization is implemented by fast Fourier transform (FFT) and the large scale least square problem is implemented by Conjugate Gradient (CG) technique, respectively. The proposed method is verified by FPGA (Xilinx Virtex-7 XC7VX690T) realization, revealing its effectiveness in real-time applications.

Download Full-text

A Machine Learning Based Strategy for Predicting the Sagittal Angles in real-time for Actuating an Active Knee-Ankle Prosthesis

10.1055/s-0040-1717439 ◽

2020 ◽

Author(s):

S Dey ◽

T Yoshida ◽

AF Schilling

Keyword(s):

Machine Learning ◽

Real Time ◽

Ankle Prosthesis

Download Full-text

Creating adaptive predictions for packaging-critical quality parameters using advanced analytics and machine learning

TAPPI Journal ◽

10.32964/tj18.11.679 ◽

2019 ◽

Vol 18 (11) ◽

pp. 679-689

Author(s):

CYDNEY RECHTIN ◽

CHITTA RANJAN ◽

ANTHONY LEWIS ◽

BETH ANN ZARKO

Keyword(s):

Machine Learning ◽

Real Time ◽

Predictive Analytics ◽

Paper Industry ◽

Quality Parameters ◽

Fast Change ◽

The Face ◽

Model Tuning ◽

Reducing Costs ◽

Time Critical

Packaging manufacturers are challenged to achieve consistent strength targets and maximize production while reducing costs through smarter fiber utilization, chemical optimization, energy reduction, and more. With innovative instrumentation readily accessible, mills are collecting vast amounts of data that provide them with ever increasing visibility into their processes. Turning this visibility into actionable insight is key to successfully exceeding customer expectations and reducing costs. Predictive analytics supported by machine learning can provide real-time quality measures that remain robust and accurate in the face of changing machine conditions. These adaptive quality “soft sensors” allow for more informed, on-the-fly process changes; fast change detection; and process control optimization without requiring periodic model tuning. The use of predictive modeling in the paper industry has increased in recent years; however, little attention has been given to packaging finished quality. The use of machine learning to maintain prediction relevancy under everchanging machine conditions is novel. In this paper, we demonstrate the process of establishing real-time, adaptive quality predictions in an industry focused on reel-to-reel quality control, and we discuss the value created through the availability and use of real-time critical quality.

Download Full-text