Denoising arterial spin labeling perfusion MRI with deep machine learning

PurposeArterial spin labeling (ASL) perfusion MRI is a noninvasive technique for measuring cerebral blood flow (CBF) in a quantitative manner. A technical challenge in ASL MRI is data processing because of the inherently low signal-to-noise-ratio (SNR). Deep learning (DL) is an emerging machine learning technique that can learn a nonlinear transform from acquired data without using any explicit hypothesis. Such a high flexibility may be particularly beneficial for ASL denoising. In this paper, we proposed and validated a DL-based ASL MRI denoising algorithm (DL-ASL).MethodsThe DL-ASL network was constructed using convolutional neural networks (CNNs) with dilated convolution and wide activation residual blocks to explicitly take the inter-voxel correlations into account, and preserve spatial resolution of input image during model learning.ResultsDL-ASL substantially improved the quality of ASL CBF in terms of SNR. Based on retrospective analyses, DL-ASL showed a high potential of reducing 75% of the original acquisition time without sacrificing CBF measurement quality.ConclusionDL-ASL achieved improved denoising performance for ASL MRI as compared with current routine methods in terms of higher PSNR, SSIM and Radiologic scores. With the help of DL-ASL, much fewer repetitions may be prescribed in ASL MRI, resulting in a great reduction of the total acquisition time.     

热成像系统图像处理电路的新进展

国外一流热成像厂商均研制了适宜其产品应用的新型高性能图像处理电路。选取了5套具有代表性的国外高性能处理电路,对其硬件设计构架、实现功能、以及对外接口三方面进行了对比分析。这些高性能处理电路的主要特点有:电路功能分模拟转换和数字处理两部分;数字处理电路以大规模FPGA为处理核心,并配以大量存储器,且广泛应用FPGA中的嵌入式核;在具备基本使用功能的基础上,继续在非均匀校正以及图像增强两个方向,应用效果更佳的算法;处理电路通常具备扩展能力;处理图像分辨率大于等于640×512像素;模拟视频采用VGA接口等。     

Pore structure analysis on activated carbon fibers—By cluster and watershed transform method

The microimage of activated carbon fibers (ACFs) obtained from SEM or TEM has low signal to noise ratio (SNR). And the objects in image overlapped together. Image processing methods based on clustering analysis and watershed transform were applied to explore the macropores morphology structure of ACFs. The microimage shows that abundant round-shaped macropores with pore size around 50-200 nm, some of which cluster in the form of ring, distribute on ACF surface. Through clustering analysis, we calculated the surface area of macropores and their distributions. The results showed that the pores characterize within three regions, (1) the pore objects smaller than 30 pixels distributing nearly uniformly; (2) those with pixels between 30 and 165 peaking at lower region and (3) those larger than 165 pixels having fewer numbers and provide the main part of clusters. According to watershed transform segmentation analysis, the overlapped pores in one cluster were separated from each other. The number and other morphology parameters of macropores are calculated automatically and accurately in this paper.     

Improving cerebral blood flow quantification for arterial spin labeled perfusion MRI by removing residual motion artifacts and global signal fluctuations

Denoising is critical to improving the quality and stability of cerebral blood flow (CBF) quantification in arterial spin labeled (ASL) perfusion magnetic resonance imaging (MRI) due to the intrinsic low signal-to-noise-ratio (SNR) of ASL data. Previous studies have been focused on reducing the spatial or temporal noise using standard filtering techniques, and less attention has been paid to two global nuisance effects, the residual motion artifacts and the global signal fluctuations. Since both nuisances affect the whole brain, removing them in advance should enhance the CBF quantification quality for ASL MRI. The purpose of this paper was to assess this potential benefit. Three methods were proposed to suppress each or both of the two global nuisances. Their performances for CBF quantification were validated using ASL data acquired from 13 subjects. Evaluation results showed that covarying out both global nuisances significantly improved temporal SNR and test-retest stability of CBF measurement. Although the concept of removing both nuisances is not technically novel per se, this paper clearly showed the benefits for ASL CBF quantification. Dissemination of the proposed methods in a free ASL data processing toolbox should be of interest to a broad range of ASL users.     

A simple and fast technique for on-line fMRI data analysis

In the present work a simple technique for fMRI data analysis is presented. Artifacts due to random and stimulus-correlated motions are corrected without image registration procedures. The first step of our procedure is the calculation of the raw activation map by correlation analysis. The task related motion artifacts arise at the tissue interfaces, including vessels: when image intensity gradient is calculated the high values correspond to interface regions. To eliminate stimulus-correlated motion artifacts the intensity gradient image, obtained from the fMRI data set, is compared to the raw activation map. Since small random motions decrease the value of the correlation coefficient (R) of the external pixels of the activation areas, in the last step of our analysis procedures the clusters are extended to connected pixels having R values smaller than the defined threshold. Each cluster is expanded until the R value of the cluster average intensity is kept constant. The procedure has been tested with both GRE and EPI studies. The presented approach is a fast and robust technique useful for preliminary or on-line analysis of fMRI data.     

Empirical optimization of ASL data analysis using an ASL data processing toolbox: ASLtbx

Arterial spin labeling (ASL) perfusion fMRI data differ in important respects from the more familiar blood oxygen level-dependent (BOLD) fMRI data and require specific processing strategies. In this paper, we examined several factors that may influence ASL data analysis, including data storage bit resolution, motion correction, preprocessing for cerebral blood flow (CBF) calculations and nuisance covariate modeling. Continuous ASL data were collected at 3 T from 10 subjects while they performed a simple sensorimotor task with an epoch length of 48 s. These data were then analyzed using systematic variations of the factors listed above to identify the approach that yielded optimal signal detection for task activation. Improvements in statistical power were found for use of at least 10 bits for data storage at 3 T. No significant difference was found in motor cortex regarding using simple subtraction or sinc subtraction, but the former presented minor but significantly (P<.024) larger peak t value in visual cortex. While artifactual head motion patterns were observed in synthetic data and background-suppressed ASL data when label/control images were realigned to a common target, independent realignment of label and control images did not yield significant improvements in activation in the sensorimotor data. It was also found that CBF calculations should be performed prior to spatial normalization and that modeling of global fluctuations yielded significantly increased peak t value in motor cortex. The implementation of all ASL data processing approaches is easily accomplished within an open-source toolbox, ASLtbx, and is advocated for most perfusion fMRI data sets.     

Automated determination of size and morphology information from soot transmission electron microscope (TEM)-generated images

The thermophoretic sampling of particulates from hot media, coupled with transmission electron microscope (TEM) imaging, is a combined approach that is widely used to derive morphological information. The identification and the measurement of the particulates, however, can be complex when the TEM images are of low contrast, noisy, and have non-uniform background signal level. The image processing method can also be challenging and time consuming, when the samples collected have large variability in shape and size, or have some degree of overlapping. In this work, a three-stage image processing sequence is presented to facilitate time-efficient automated identification and measurement of particulates from the TEM grids. The proposed processing sequence is first applied to soot samples that were thermophoretically sampled from a laminar non-premixed ethylene-air flame. The parameter values that are required to be set to facilitate the automated process are identified, and sensitivity of the results to these parameters is assessed. The same analysis process is also applied to soot samples that were acquired from an externally irradiated laminar non-premixed ethylene-air flame, which have different geometrical characteristics, to assess the morphological dependence of the proposed image processing sequence. Using the optimized parameter values, statistical assessments of the automated results reveal that the largest discrepancies that are associated with the estimated values of primary particle diameter, fractal dimension, and prefactor values of the aggregates for the tested cases, are approximately 3, 1, and 10 %, respectively, when compared with the manual measurements.     

Robust estimation of the cerebral blood flow in arterial spin labelling

The introduction of arterial spin labelling (ASL) techniques in magnetic resonance imaging (MRI) has made feasible a non-invasive measurement of the cerebral blood flow (CBF). However, to date, the low signal-to-noise ratio of ASL gives us no option but to repeat the acquisition to accumulate enough data in order to get a reliable signal. The perfusion signal is then usually extracted by averaging across the repetitions. But the sample mean is very sensitive to outliers. A single incorrect observation can therefore be the source of strong detrimental effects on the perfusion-weighted image estimated with the sample mean.     

数字散斑识别算法中的GPU高性能运算应用研究

数字散斑相关方法有着测量环境简单、全场非接触等优点,但算法效率一直是限制其发展的瓶颈之一。GPU有着天然的并行性,GPU高性能运算可以为计算机图形处理带来极大的效率提升。利用CUDA平台编程对传统的数字散斑逐点搜索算法、十字搜索算法及遗传算法进行GPU高性能并行处理,并与传统方法比较分析。实验结果表明,对于尺寸为150150像素的散斑图像,3种方法效率分别提升了20倍、8倍、31倍;对于尺寸为500500像素的散斑图像,3种方法效率分别提升了183倍、33倍、44倍;对于尺寸为1 0001 000像素的散斑图像,3种方法效率分别提升了424倍、116倍、44倍。     

基于图像处理的变电站中隔离开关的状态研究

随着图像处理技术的不断发展,本文利用图像处理技术分析变电站中隔离开关的状态。蚁群算法(Ant Colony Algorithm,ACA)使用的广泛性,很多学者将其应用到图像处理中。本文将蚁群算法应用于变电站设备区域图像分割中,从某个或某些像素点出发,提取出变电站的隔离开关信息,然后对其进一步的图像处理,分析隔离开关的状态。但是,蚁群算法在运算过程中,易出现过早收敛于局部最优解及运算时间过长的缺点。为了使蚁群算法收敛于全局最优解及加快收敛速度,本文针对传统的蚁群算法模型对其信息浓度更新规则改进及参数的改进。通过仿真对比分析改进后的蚁群算法对于图像分割效果更好。     

基于嵌入式GPU的红外弱小目标检测算法

红外弱小目标的目标像素少,目标对比度低,成像帧率高,图像数据量大,检测实时性强。针对红外弱小目标检测算法适合于GPU并行计算的特点,对其在嵌入式GPU平台Jetson TX2上进行了并行优化实现。在检测算法设计、内存访问、调试优化3个方面进行了优化设计。实验结果表明,对640×480像素分辨率的红外视频,并行优化后的目标检测算法能够在10 ms内完成计算,满足实时处理需求。     

Automated optical analysis of young's fringes— optical autocorrelation

Extraction of displacement or velocity data from laser speckle photography or particle image velocimetry transparencies requires measurement of the spacing and orientation of several thousand Young's fringe patterns. In contrast with classical interferometry, fringes formed in this way have poor visibility and it is generally desirable to compute the two-dimensional Fourier transform in order to extract the fringe frequency. The large amount of data produced makes digital acquisition and image processing computationally intensive and time consuming. This paper introduces an automated processor which utilises a photorefractive crystal of bismuth silicon oxide to perform this analysis in a fast and cost-effective manner.     

Reliability and reproducibility of perfusion MRI in cognitively normal subjects

Arterial spin labeling (ASL) magnetic resonance imaging (MRI) is becoming a popular method for measuring perfusion due to its ability of generating perfusion maps noninvasively. This allows for frequent repeat scanning, which is especially useful for follow-up studies. However, limited information is available regarding the reliability and reproducibility of ASL perfusion measurements. Here, the reliability and reproducibility of pulsed ASL was investigated in an elderly population to determine the variation in perfusion among cognitively normal individuals in different brain structures. Intraclass correlation coefficients (ICC) and within-subject variation coefficients (wsCV) were used to estimate reliability and reproducibility over a period of 1 year. Twelve cognitively normal subjects (75.5±5.3 years old, six male and six female) were scanned four times (at 0, 3, 6 and 12 months). No significant difference in cerebral blood flow (CBF) was found over this period. CBF values ranged from 46 to 53 ml/100 g per minute in the medial frontal gyrus (MFG) and from 40 to 44 ml/100 g per minute over all gray matter regions in the superior part of the brain. Data obtained from the first two scans were processed by two readers and showed high reliability (ICC >0.97) and reproducibility (wsCV <6%). However, over the total period of 1 year, reliability reduced to a moderate level (ICC=0.63–0.74) with wsCVs of gray matter, left MFG, right MFG of 13.5%, 12.3%, and 15.4%, respectively. In conclusion, measurement of CBF with pulsed ASL provided good agreement between inter-raters. A moderate level of reliability was obtained over a 1-year period, which was attributed to variance in slice positioning and coregistration. As such pulsed ASL has the potential to be used for CBF comparison in longitudinal studies.     

基于改进暗目标法山区HJ CCD影像气溶胶光学厚度反演

国产HJ CCD影像在环境与灾害监测预报方面具有巨大的应用潜力,但其缺少中红外波段无法直接应用传统暗目标法反演气溶胶光学厚度,尤其山区受地形影响,气溶胶时空分布变化显著,影响了HJ CCD影像的大气校正精度。基于山区森林植被分布广的特点引入红波段直方图阈值法自动提取山区浓密植被暗像元,构建红蓝波段地表反射率间的线性关系,利用6S辐射传输模型建立查找表反演暗像元的气溶胶光学厚度,并通过空间插值推演到整幅影像。HJ CCD影像反演结果与MODIS气溶胶产品的空间分布趋势一致性非常好,且前者具有更高的空间分辨率特性,更适合山区气溶胶的遥感监测。二者的散点图拟合曲线为y=0.828 6x-0.001,R2达到0.984 3,表明改进暗目标法能有效地反演山区HJ CCD影像气溶胶光学厚度。工作中对传统暗目标法的改进有效地解决了HJ CCD影像只有可见光、近红外波段,在求解辐射传输方程时信息不足的问题,同时,改进方法中充分考虑了山区复杂地形环境的影响,为山区HJ CCD影像进行逐点大气校正奠定了基础,并为其自动化处理提供了可能。分析还显示,红波段直方图阈值法提取浓密森林暗像元比NDVI阈值法具有明显的优势;查找表建立和暗像元红、蓝波段地表反射率关系都会显著的影响气溶胶光学厚度的反演精度,是进一步改善算法、提高反演精度的重要研究方向。     

Measuring extraocular muscle volume using dynamic contours

The effect of medical treatment on extraocular muscle enlargement in thyroid associated ophthalmopathy (TAO) may be monitored by measuring the change in volume of the extraocular muscles on serial orbital MRI examinations. In theory, 3D image sets offer the opportunity to minimise errors due to poor repositioning and partial volume effects. This study describes an automated technique for estimating extraocular muscle volumes from 3D datasets. Operator input is minimal and the technique is robust. Verification of the technique on both simulated and real datasets is described. For simulated image sets, both automated segmentation and manual outlining produced estimates of volume which were on average 4% less than "true" volume. For real patient data, extraocular muscle volumes measured by the automated technique were 1.6% (SD 13%) less than volumes measured by manual outlining. Coefficient of variation for repeat outlining of the same image dataset for the automated technique was 1.0%, compared with 4% for manual outlining. The manual technique took an experienced operator approximately 20 min to perform, compared to 7 min for the automated technique. The automated method is therefore rapid, reproducible and at least as accurate as other available methods.     

Inversion profiles of adiabatic inversion pulses for flowing spins: the effects on labeling efficiency and labeling accuracy in perfusion imaging with pulsed arterial spin-labeling

The inversion profile of adiabatic inversion pulses is essential to the accuracy of perfusion measurement with pulsed arterial spin-labeling (ASL). In this paper, the inversion profiles for flowing spins were investigated using a numerical solution of the modified Bloch equations including a term for moving spins. Inversion profiles for spins flowing at a constant or varying velocity were examined for hyperbolic secant (HS) and frequency-offset corrected inversion (FOCI) pulses. Distortions of the inversion profiles were found for both pulses with spins flowing within physiological velocity range. The effects of the distorted profiles on labeling efficiency and labeling accuracy in the application of pulsed ASL perfusion imaging were analyzed. These effects should be taken into account in ASL techniques, in order to obtain robust and accurate perfusion measurements.     

Multi-frame sub-pixel processing algorithm based on uncontrolled micro-scanning

For the uncontrolled micro-scanning where the displacement of image sequence is random, a sub-pixel processing algorithm based on boundary recursive and subsequent error optimization is proposed in this paper. Neighboring pixels are used to approximate the boundary pixels of the original image in the algorithm. Then the approximate error is optimized by the gray statistical principle. At last a sub-pixel image with high resolution is reconstructed. The results show that our algorithm is better than the general over-sample reconstruction algorithms and robust in different displacement situations.     

基于相位差混合处理方法的高分辨力成像技术

为了提高成像系统的分辨能力, 并尽量减小系统的复杂度, 本文将相位差波前探测技术和相位差图像恢复技术结合起来构成相位差混合处理方法, 给出了点目标和扩展目标情况下混合处理方法的数值仿真结果, 并针对点目标情况进行了实验验证. 实验表明, 在像差较大的情况下, 直接用事后处理方法无法得到满意的结果. 在三种湍流强度下, 经混合方法处理后得到光斑的半高宽分别由自适应光学系统校正后的5.1, 5.1和5.0个像素减小到3.3, 3.2和3.0个像素. 可以看出, 利用相位差混合处理方法得到的图像明显优于单独的事后图像处理方法和自适应光学校正, 相位差混合处理方法在高分辨力成像领域有着巨大的应用潜力.     

Fast transforms for speckle photography fringe analysis

Two image processing techniques for automatic analysis of the Young's fringes diffraction patterns from a double-exposure speckle photograph are described. The techniques involve two-dimensional spectral analysis of the image using the Fast Fourier Transform (FFT) or the Fast Walsh Transform (FWT). The accuracy and reliability of the two techniques are compared as a function of the fringe visibility, and the number of pixels used to represent the image. Walsh analysis is found to offer the same accuracy as Fourier analysis, but in about one third the computation time. The Fourier method is more reliable with fringes of the lowest visibility, however. The ideas presented have relevance to the general problem of detecting and accurately determining the frequency components of a sinusoid in the presence of noise.     

气粒并流上行流动中颗粒团特性的实验研究

用流动显示法和图像分析法研究了气粒并流上行流动中颗粒团的瞬态行为,通过灰度梯度识别颗粒团,由颗粒团成像的象素确定其尺寸,采用相关分析法计算颗粒团的速度。随着颗粒流量的提高和表现气速的降低,气粒并流上行流动中开始出现颗粒团,由于颗粒团与气相的相互作用,流态具有不稳定和不规则的现象.在一定操作条件下,在壁面附近出现了一些长时间停留的大尺寸颗粒团.