The CUBLAS and CULA based GPU acceleration of adaptive finite element framework for bioluminescence tomography

Embedded parallel Fourier ptychographic microscopy reconstruction system

Fourier ptychographic microscopy (FPM) has attracted a wide range of focus for its ability of large space-bandwidth product and quantitative phase imaging. It is a typical computational imaging technique that jointly optimizes imaging hardware and reconstruction algorithms. The data redundancy and inverse problem algorithms are the sources of FPM's excellent performance. But at the same time, this large amount of data processing and complex algorithms also evidently reduce the imaging speed. To accelerate the FPM reconstruction speed, we proposed a fast FPM reconstruction framework consisting of three levels of parallel computation and implemented it with an embedded computing module. In the conventional FPM framework, the sample image is divided into multiple sub-regions to process separately because the illumination angles and defocus distances for different sub-regions may also be different. Our parallel framework first performs digital refocusing and high-resolution reconstruction for each sub-region separately and then stitches the complex sub-regions together to obtain the final high-resolution complex image. The feasibility of the proposed parallel FPM reconstruction framework is verified with different experimental results acquired with the system we built.

Fast direct reconstruction strategy of dynamic fluorescence molecular tomography using graphics processing units

Dynamic fluorescence molecular tomography (DFMT) is a valuable method to evaluate the metabolic process of contrast agents in different organs in vivo, and direct reconstruction methods can improve the temporal resolution of DFMT. However, challenges still remain due to the large time consumption of the direct reconstruction methods. An acceleration strategy using graphics processing units (GPU) is presented. The procedure of conjugate gradient optimization in the direct reconstruction method is programmed using the compute unified device architecture and then accelerated on GPU. Numerical simulations and in vivo experiments are performed to validate the feasibility of the strategy. The results demonstrate that, compared with the traditional method, the proposed strategy can reduce the time consumption by ∼90% without a degradation of quality.

Enhanced immunotherapy of SM5-1 in hepatocellular carcinoma by conjugating with gold nanoparticles and its in vivo bioluminescence tomographic evaluation

SM5-1 is a humanized mouse monoclonal antibody, targeting an over-expressed membrane protein of approximately 230 kDa in hepatocellular carcinoma (HCC). SM5-1 can be used for target therapy in hepatocellular carinoma due to its ability of inhibiting cell growth and inducing apoptosis. However, the tumor inhibition efficacy of SM5-1 in HCC cancer treatment remains low. In this study, we synthesized SM5-1-conjugated gold nanoparticles (Au-SM5-1 NPs) and investigated their anticancer efficacy in HCC both in vitro and in vivo. The tumor inhibition rates of Au-SM5-1 NPs for subcutaneous tumor mice were 40.10% ± 4.34%, 31.37% ± 5.12%, and 30.63% ± 4.87% on day 12, 18, and 24 post-treatment as determined by bioluminescent intensity. In addition, we investigated the antitumor efficacy of Au-SM5-1 NPs in orthotopic HCC tumor models. The results showed that the inhibition rates of Au-SM5-1 NPs can reach up to 39.64% ± 4.87% on day 31 post-treatment determined by the bioluminescent intensity of the abdomen in tumor-bearing mice. Furthermore, three-dimensional reconstruction results of the orthotopic tumor revealed that Au-SM5-1 NPs significantly inhibited tumor growth compared with SM5-1 alone. Our results suggested that the developed Au-SM5-1 NPs has great potential as an antibody-based nano-drug for HCC therapy.

Reconstruction of fluorophore concentration variation in dynamic fluorescence molecular tomography

Dynamic fluorescence molecular tomography (DFMT) is a potential approach for drug delivery, tumor detection, diagnosis, and staging. The purpose of DFMT is to quantify the changes of fluorescent agents in the bodies, which offer important information about the underlying physiological processes. However, the conventional method requires that the fluorophore concentrations to be reconstructed are stationary during the data collection period. As thus, it cannot offer the dynamic information of fluorophore concentration variation within the data collection period. In this paper, a method is proposed to reconstruct the fluorophore concentration variation instead of the fluorophore concentration through a linear approximation. The fluorophore concentration variation rate is introduced by the linear approximation as a new unknown term to be reconstructed and is used to obtain the time courses of fluorophore concentration. Simulation and phantom studies are performed to validate the proposed method. The results show that the method is able to reconstruct the fluorophore concentration variation rates and the time courses of fluorophore concentration with relative errors less than 0.0218.

GPU-based block-wise nonlocal means denoising for 3D ultrasound images

Speckle suppression plays an important role in improving ultrasound (US) image quality. While lots of algorithms have been proposed for 2D US image denoising with remarkable filtering quality, there is relatively less work done on 3D ultrasound speckle suppression, where the whole volume data rather than just one frame needs to be considered. Then, the most crucial problem with 3D US denoising is that the computational complexity increases tremendously. The nonlocal means (NLM) provides an effective method for speckle suppression in US images. In this paper, a programmable graphic-processor-unit- (GPU-) based fast NLM filter is proposed for 3D ultrasound speckle reduction. A Gamma distribution noise model, which is able to reliably capture image statistics for Log-compressed ultrasound images, was used for the 3D block-wise NLM filter on basis of Bayesian framework. The most significant aspect of our method was the adopting of powerful data-parallel computing capability of GPU to improve the overall efficiency. Experimental results demonstrate that the proposed method can enormously accelerate the algorithm.

Toward real-time availability of 3D temperature maps created with temporally constrained reconstruction

PURPOSE

To extend the previously developed temporally constrained reconstruction (TCR) algorithm to allow for real-time availability of three-dimensional (3D) temperature maps capable of monitoring MR-guided high intensity focused ultrasound applications.

METHODS

A real-time TCR (RT-TCR) algorithm is developed that only uses current and previously acquired undersampled k-space data from a 3D segmented EPI pulse sequence, with the image reconstruction done in a graphics processing unit implementation to overcome computation burden. Simulated and experimental data sets of HIFU heating are used to evaluate the performance of the RT-TCR algorithm.

RESULTS

The simulation studies demonstrate that the RT-TCR algorithm has subsecond reconstruction time and can accurately measure HIFU-induced temperature rises of 20°C in 15 s for 3D volumes of 16 slices (RMSE = 0.1°C), 24 slices (RMSE = 0.2°C), and 32 slices (RMSE = 0.3°C). Experimental results in ex vivo porcine muscle demonstrate that the RT-TCR approach can reconstruct temperature maps with 192 × 162 × 66 mm 3D volume coverage, 1.5 × 1.5 × 3.0 mm resolution, and 1.2-s scan time with an accuracy of ±0.5°C.

CONCLUSION

The RT-TCR algorithm offers an approach to obtaining large coverage 3D temperature maps in real-time for monitoring MR-guided high intensity focused ultrasound treatments.

Acceleration of early-photon fluorescence molecular tomography with graphics processing units

Fluorescence molecular tomography (FMT) with early-photons can improve the spatial resolution and fidelity of the reconstructed results. However, its computing scale is always large which limits its applications. In this paper, we introduced an acceleration strategy for the early-photon FMT with graphics processing units (GPUs). According to the procedure, the whole solution of FMT was divided into several modules and the time consumption for each module is studied. In this strategy, two most time consuming modules (G_d and W modules) were accelerated with GPU, respectively, while the other modules remained coded in the Matlab. Several simulation studies with a heterogeneous digital mouse atlas were performed to confirm the performance of the acceleration strategy. The results confirmed the feasibility of the strategy and showed that the processing speed was improved significantly.

Fluorescence molecular tomography using a two-step three-dimensional shape-based reconstruction with graphics processing unit acceleration

In fluorescence molecular tomography, the accurate and stable reconstruction of fluorescence-labeled targets remains a challenge for wide application of this imaging modality. Here we propose a two-step three-dimensional shape-based reconstruction method using graphics processing unit (GPU) acceleration. In this method, the fluorophore distribution is assumed as the sum of ellipsoids with piecewise-constant fluorescence intensities. The inverse problem is formulated as a constrained nonlinear least-squares problem with respect to shape parameters, leading to much less ill-posedness as the number of unknowns is greatly reduced. Considering that various shape parameters contribute differently to the boundary measurements, we use a two-step optimization algorithm to handle them in a distinctive way and also stabilize the reconstruction. Additionally, the GPU acceleration is employed for finite-element-method-based calculation of the objective function value and the Jacobian matrix, which reduces the total optimization time from around 10 min to less than 1 min. The numerical simulations show that our method can accurately reconstruct multiple targets of various shapes while the conventional voxel-based reconstruction cannot separate the nearby targets. Moreover, the two-step optimization can tolerate different initial values in the existence of noises, even when the number of targets is not known a priori. A physical phantom experiment further demonstrates the method's potential in practical applications.

High-performance image reconstruction in fluorescence tomography on desktop computers and graphics hardware

Image reconstruction in fluorescence optical tomography is a three-dimensional nonlinear ill-posed problem governed by a system of partial differential equations. In this paper we demonstrate that a combination of state of the art numerical algorithms and a careful hardware optimized implementation allows to solve this large-scale inverse problem in a few seconds on standard desktop PCs with modern graphics hardware. In particular, we present methods to solve not only the forward but also the non-linear inverse problem by massively parallel programming on graphics processors. A comparison of optimized CPU and GPU implementations shows that the reconstruction can be accelerated by factors of about 15 through the use of the graphics hardware without compromising the accuracy in the reconstructed images.

GPU-Accelerated Finite Element Method for Modelling Light Transport in Diffuse Optical Tomography

We introduce a GPU-accelerated finite element forward solver for the computation of light transport in scattering media. The forward model is the computationally most expensive component of iterative methods for image reconstruction in diffuse optical tomography, and performance optimisation of the forward solver is therefore crucial for improving the efficiency of the solution of the inverse problem. The GPU forward solver uses a CUDA implementation that evaluates on the graphics hardware the sparse linear system arising in the finite element formulation of the diffusion equation. We present solutions for both time-domain and frequency-domain problems. A comparison with a CPU-based implementation shows significant performance gains of the graphics accelerated solution, with improvements of approximately a factor of 10 for double-precision computations, and factors beyond 20 for single-precision computations. The gains are also shown to be dependent on the mesh complexity, where the largest gains are achieved for high mesh resolutions.