Search NASA⌕ Search

Engineering topics

Li, L

Publications and source records attributed to Li, L.

At least 19 records

Scalable Deep Learning-Based Microarchitecture Simulation on GPUs

Cycle-accurate microarchitecture simulators are essential tools for designers to architect, estimate, optimize, and manufacture new processors that meet specific design expectations. However, conventional simulators based on discrete-event methods often require an exceedingly long time-to-solution for the simulation of applications and architectures at full complexity and scale. Given the excitement around wielding the machine learning (ML) hammer to tackle various architecture problems, there have been attempts to employ ML to perform architecture simulations, such as Ithemal and SimNet. However, the direct application of existing ML approaches to architecture simulation may be even slower due to overwhelming memory traffic and stringent sequential computation logic. This work proposes the first graphics processing unit (GPU)-based microarchitecture simulator that fully unleashes the potential of GPUs to accelerate state-of-the-art ML-based simulators. First, considering the application traces are loaded from central processing unit (CPU) to GPU for simulation, we introduce various designs to reduce the data movement cost between CPUs and GPUs. Second, we propose a parallel simulation paradigm that partitions the application trace into sub-traces to simulate them in parallel with rigorous error analysis and effective error correction mechanisms. Combined, this scalable GPU-based simulator outperforms by orders of magnitude the traditional CPU-based simulators and the state-of-the-art ML-based simulators, i.e., SimNet and Ithemal.

97 MATHEMATICS AND COMPUTING↗

C-SAW: a framework for graph sampling and random walk on GPUs

Many applications require to learn, mine, analyze and visualize large-scale graphs. These graphs are often too large to be addressed efficiently using conventional graph processing technologies. Fortunately, recent research efforts find out graph sampling and random walk, which significantly reduce the size of original graphs, can benefit the tasks of learning, mining, analyzing and visualizing large graphs by capturing the desirable graph properties. This paper introduces C-SAW, the first framework that accelerates Sampling and Random Walk framework on GPUs. Particularly, C-SAW makes three contributions: First, our framework provides a generic API which allows users to implement a wide range of sampling and random walk algorithms with ease. Second, offloading this framework on GPU, we introduce warp-centric parallel selection, and two novel optimizations for collision migration. Third, towards supporting graphs that exceed the GPU memory capacity, we introduce efficient data transfer optimizations for out-of-memory and multi-GPU sampling, such as workload-aware scheduling and batched multi-instance sampling. Taken together, our framework constantly outperforms the state of the art projects in addition to the capability of supporting a wide range of sampling and random walk algorithms.

97 MATHEMATICS AND COMPUTING↗

TU-H-CAMPUS-IeP3-01: Simultaneous PET Restoration and PET/CT Co-Segmentation Using a Variational Method

Purpose: PET images are usually blurred due to the finite spatial resolution, while CT images suffer from low contrast. Segment a tumor from either a single PET or CT image is thus challenging. To make full use of the complementary information between PET and CT, we propose a novel variational method for simultaneous PET image restoration and PET/CT images co-segmentation. Methods: The proposed model was constructed based on the Γ-convergence approximation of Mumford-Shah (MS) segmentation model for PET/CT co-segmentation. Moreover, a PET de-blur process was integrated into the MS model to improve the segmentation accuracy. An interaction edge constraint term over the two modalities were specially designed to share the complementary information. The energy functional was iteratively optimized using an alternate minimization (AM) algorithm. The performance of the proposed method was validated on ten lung cancer cases and five esophageal cancer cases. The ground truth were manually delineated by an experienced radiation oncologist using the complementary visual features of PET and CT. The segmentation accuracy was evaluated by Dice similarity index (DSI) and volume error (VE). Results: The proposed method achieved an expected restoration result for PET image and satisfactory segmentation results for both PET and CT images. For lung cancer dataset, the average DSI (0.72) increased by 0.17 and 0.40 than single PET and CT segmentation. For esophageal cancer dataset, the average DSI (0.85) increased by 0.07 and 0.43 than single PET and CT segmentation. Conclusion: The proposed method took full advantage of the complementary information from PET and CT images. This work was supported in part by the National Cancer Institute Grants R01CA172638. Shan Tan and Laquan Li were supported in part by the National Natural Science Foundation of China, under Grant Nos. 60971112 and 61375018.

60 APPLIED LIFE SCIENCES↗

TH-CD-206-01: Expectation-Maximization Algorithm-Based Tissue Mixture Quantification for Perfusion MRI

Purpose: To investigate the feasibility of estimating the tissue mixture perfusions and quantifying cerebral blood flow change in arterial spin labeled (ASL) perfusion MR images. Methods: The proposed perfusion MR image analysis framework consists of 5 steps: (1) Inhomogeneity correction was performed on the T1- and T2-weighted images, which are available for each studied perfusion MR dataset. (2) We used the publicly available FSL toolbox to strip off the non-brain structures from the T1- and T2-weighted MR images. (3) We applied a multi-spectral tissue-mixture segmentation algorithm on both T1- and T2-structural MR images to roughly estimate the fraction of each tissue type - white matter, grey matter and cerebral spinal fluid inside each image voxel. (4) The distributions of the three tissue types or tissue mixture across the structural image array are down-sampled and mapped onto the ASL voxel array via a co-registration operation. (5) The presented 4-dimensional expectation-maximization (4D-EM) algorithm takes the down-sampled three tissue type distributions on perfusion image data to generate the perfusion mean, variance and percentage images for each tissue type of interest. Results: Experimental results on three volunteer datasets demonstrated that the multi-spectral tissue-mixture segmentation algorithm was effective to initialize tissue mixtures from T1- and T2-weighted MR images. Compared with the conventional ASL image processing toolbox, the proposed 4D-EM algorithm not only generated comparable perfusion mean images, but also produced perfusion variance and percentage images, which the ASL toolbox cannot obtain. It is observed that the perfusion contribution percentages may not be the same as the corresponding tissue mixture volume fractions estimated in the structural images. Conclusion: A specific application to brain ASL images showed that the presented perfusion image analysis method is promising for detecting subtle changes in tissue perfusions, which is valuable for the early diagnosis of certain brain diseases, e.g. multiple sclerosis.

60 APPLIED LIFE SCIENCES↗

Measurement of the decay B → D ℓ ν ℓ in fully reconstructed events and determination of the Cabibbo-Kobayashi-Maskawa matrix element | V c b |

We present a determination of the magnitude of the Cabibbo-Kobayashi-Maskawa matrix element |Vcb| using the decay B → Dℓν ℓ (ℓ = e,μ) based on 711 fb –1 of e + e – → Υ(4S) data recorded by the Belle detector and containing 772 × 10 6 BB¯ pairs. Here, one B meson in the event is fully reconstructed in a hadronic decay mode, while the other, on the signal side, is partially reconstructed from a charged lepton and either a D + or D 0 meson in a total of 23 hadronic decay modes. The isospin-averaged branching fraction of the decay B → Dℓν ℓ is found to be B(B 0 → D – ℓ + ν ℓ ) = (2.31 ± 0.03(stat) ± 0.11(syst))%. Analyzing the differential decay rate as a function of the hadronic recoil with the parametrization of Caprini, Lellouch, and Neubert and using the form-factor prediction G(1) = 1.0541 ± 0.0083 calculated by FNAL/MILC, we obtain η EW |V cb | = (40.12 ± 1.34) × 10 –3 , where ηEW is the electroweak correction factor. Alternatively, assuming the model-independent form-factor parametrization of Boyd, Grinstein, and Lebed and using lattice QCD data from the FNAL/MILC and HPQCD collaborations, we find η EW |V cb | = (41.10 ± 1.14) × 10 –3 .

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

First observation of the decay B 0 → ψ ( 2 S ) π 0

We report a measurement of the B 0 → ψ(2S)π 0 branching fraction based on the full Υ(4S) data set of 772 × 10 6 BB¯ pairs collected by the Belle detector at the KEKB asymmetric-energy e + e – collider. We obtain B(B 0 → ψ(2S)π 0 ) = (1.17 ± 0.17(stat) ± 0.08(syst)) × 10 –5 . Here, the result has a significance of 7.2 standard deviations and is the first observation of the decay B 0 → ψ(2S)π 0 .

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

TU-CD-BRA-12: Coupling PET Image Restoration and Segmentation Using Variational Method with Multiple Regularizations

Purpose: To propose a new variational method which couples image restoration with tumor segmentation for PET images using multiple regularizations. Methods: Partial volume effect (PVE) is a major degrading factor impacting tumor segmentation accuracy in PET imaging. The existing segmentation methods usually need to take prior calibrations to compensate PVE and they are highly system-dependent. Taking into account that image restoration and segmentation can promote each other and they are tightly coupled, we proposed a variational method to solve the two problems together. Our method integrated total variation (TV) semi-blind deconvolution and Mumford-Shah (MS) segmentation. The TV norm was used on edges to protect the edge information, and the L{sub 2} norm was used to avoid staircase effect in the no-edge area. The blur kernel was constrained to the Gaussian model parameterized by its variance and we assumed that the variances in the X-Y and Z directions are different. The energy functional was iteratively optimized by an alternate minimization algorithm. Segmentation performance was tested on eleven patients with non-Hodgkin’s lymphoma, and evaluated by Dice similarity index (DSI) and classification error (CE). For comparison, seven other widely used methods were also tested and evaluated. Results: The combination of TV and L{sub 2} regularizations effectively improved the segmentation accuracy. The average DSI increased by around 0.1 than using either the TV or the L{sub 2} norm. The proposed method was obviously superior to other tested methods. It has an average DSI and CE of 0.80 and 0.41, while the FCM method — the second best one — has only an average DSI and CE of 0.66 and 0.64. Conclusion: Coupling image restoration and segmentation can handle PVE and thus improves tumor segmentation accuracy in PET. Alternate use of TV and L2 regularizations can further improve the performance of the algorithm. This work was supported in part by National Natural Science Foundation of China (NNSFC), under Grant No.61375018, and Fundamental Research Funds for the Central Universities, under Grant No. 2012QN086. Wei Lu was supported in part by the National Institutes of Health (NIH) Grant No. R01 CA172638.

60 APPLIED LIFE SCIENCES↗

SU-C-9A-03: Simultaneous Deconvolution and Segmentation for PET Tumor Delineation Using a Variational Method

Purpose: To implement a new method that integrates deconvolution with segmentation under the variational framework for PET tumor delineation. Methods: Deconvolution and segmentation are both challenging problems in image processing. The partial volume effect (PVE) makes tumor boundaries in PET image blurred which affects the accuracy of tumor segmentation. Deconvolution aims to obtain a PVE-free image, which can help to improve the segmentation accuracy. Conversely, a correct localization of the object boundaries is helpful to estimate the blur kernel, and thus assist in the deconvolution. In this study, we proposed to solve the two problems simultaneously using a variational method so that they can benefit each other. The energy functional consists of a fidelity term and a regularization term, and the blur kernel was limited to be the isotropic Gaussian kernel. We minimized the energy functional by solving the associated Euler-Lagrange equations and taking the derivative with respect to the parameters of the kernel function. An alternate minimization method was used to iterate between segmentation, deconvolution and blur-kernel recovery. The performance of the proposed method was tested on clinic PET images of patients with non-Hodgkin's lymphoma, and compared with seven other segmentation methods using the dice similarity index (DSI) and volume error (VE). Results: Among all segmentation methods, the proposed one (DSI=0.81, VE=0.05) has the highest accuracy, followed by the active contours without edges (DSI=0.81, VE=0.25), while other methods including the Graph Cut and the Mumford-Shah (MS) method have lower accuracy. A visual inspection shows that the proposed method localizes the real tumor contour very well. Conclusion: The result showed that deconvolution and segmentation can contribute to each other. The proposed variational method solve the two problems simultaneously, and leads to a high performance for tumor segmentation in PET. This work was supported in part by National Natural Science Foundation of China (NNSFC), under Grant Nos. 60971112 and 61375018, and Fundamental Research Funds for the Central Universities, under Grant No. 2012QN086. Wei Lu was supported in part by the National Institutes of Health (NIH) Grant No. R01 CA172638.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

SU-E-I-96: A Study About the Influence of ROI Variation On Tumor Segmentation in PET

Purpose: To study the influence of different regions of interest (ROI) on tumor segmentation in PET. Methods: The experiments were conducted on a cylindrical phantom. Six spheres with different volumes (0.5ml, 1ml, 6ml, 12ml, 16ml and 20 ml) were placed inside a cylindrical container to mimic tumors of different sizes. The spheres were filled with 11C solution as sources and the cylindrical container was filled with 18F-FDG solution as the background. The phantom was continuously scanned in a Biograph-40 True Point/True View PET/CT scanner, and 42 images were reconstructed with source-to-background ratio (SBR) ranging from 16:1 to 1.8:1. We took a large and a small ROI for each sphere, both of which contain the whole sphere and does not contain any other spheres. Six other ROIs of different sizes were then taken between the large and the small ROI. For each ROI, all images were segmented by eitht thresholding methods and eight advanced methods, respectively. The segmentation results were evaluated by dice similarity index (DSI), classification error (CE) and volume error (VE). The robustness of different methods to ROI variation was quantified using the interrun variation and a generalized Cohen's kappa. Results: With the change of ROI, the segmentation results of all tested methods changed more or less. Compared with all advanced methods, thresholding methods were less affected by the ROI change. In addition, most of the thresholding methods got more accurate segmentation results for all sphere sizes. Conclusion: The results showed that the segmentation performance of all tested methods was affected by the change of ROI. Thresholding methods were more robust to this change and they can segment the PET image more accurately. This work was supported in part by National Natural Science Foundation of China (NNSFC), under Grant Nos. 60971112 and 61375018, and Fundamental Research Funds for the Central Universities, under Grant No. 2012QN086. Wei Lu was supported in part by the National Institutes of Health (NIH) Grant No. R01 CA172638.

60 APPLIED LIFE SCIENCES↗