基于CUDA的快速三维医学图像分割

目的:三维分割是医学图像分析和可视化中的重要组成部分,也是医学图像分割中的一个难点。水平集方法在三维医学图像分割中有很广阔的应用前景,但是该算法的计算量大,不能达到实时处理的要求。针对这个问题,提出了一种基于CUDA的并行加速方法。方法:采用NVIDIA公司的GPGPU模型CUDA,利用图像像素的独立性和偏微分方程求解的并发性,提高C-V水平集算法的分割速度。给出了并行计算的流程图,并对C-V水平集算法在CUDA上的实现进行了详细介绍。结果:实现了C-V水平集并行加速算法,该方法在保证分割效果的前提下,具有更快的分割速度。结论:所提出的方法是切实可行的,实现了快速的三维医学图像分割。     

The influence of rTMS over the right dorsolateral prefrontal cortex on intentional set switching

High frequency (HF) repetitive transcranial magnetic stimulation (rTMS) has an excitatory effect on neurons of a specific brain area. The dorsolateral prefrontal cortex (DLPFC) has been associated with executive functions, such as task set switching. One important experimental paradigm for investigating such higher order cognitive control is the task-switching (TS) paradigm. A TS paradigm requires switching between two conditional response tasks with mutually incompatible response–selection rules. In the present study, the influence of HF rTMS over the right DLPFC in healthy female volunteers on a modified TS paradigm was investigated. As expected, reaction time on cued switching trials decreased significant after rTMS, as compared to non-cued switch trials. No changes emerged after the placebo sham condition. Mood remained unchanged after rTMS. These findings demonstrate the role of the right DLPFC in cued intentional set switch initiation.     

Practical tips and tricks for the measurement of fractional flow reserve

Coronary pressure measurement has emerged as one of the most promising tool to assist interventional cardiologist to make accurate clinical decisions. Fractional flow reserve (FFR) is a simple way of reliably assessing the functional consequences of coronary atherosclerosis. The training in FFR is easy, and coronary pressure measurements can be obtained swiftly in the catheterization laboratory. This review outlines the practical set up of FFR in the catheterization laboratory and also discusses the potential pitfalls in the measurement of FFR. © 2010 Wiley‐Liss, Inc.     

Prestimulus EEG gamma frequencies during formation of a cognitive set to facial expressions

The power spectra of baseline (prestimulus) EEG recordings from various parts of the cortex in the band 1–60 Hz were studied in adult subjects during the stage of formation of a cognitive set to facial expressions. Individual variability in the power spectra of the baseline EEG, particularly in the gamma range (41–60 Hz) made it impossible to average spectra from different subjects. This is the first study demonstrating that the EEG at 41–60 Hz was more informative during the prestimulus period than recordings at 1–20 Hz and 21–40 Hz. High levels of gamma-frequency power were found to be dominant in the frontal areas of the right hemisphere in groups of subjects with plastic sets; the group of subjects with a rigid set showed dominance of gamma-frequency power in the posterior temporal and occipital areas of the left hemisphere. Translated from Zhurnal Vysshei Nervnoi Deyatel'nosti imeni I. P. Pavlova, Vol. 58, No. 1, pp. 5–11, January–February, 2008.     

Embryonic stem cell gene expression signatures in the canine mammary tumor: a bioinformatics approach

Canine breast cancer was considered as an ideal model of comparative oncology for the human breast cancer, as there is significant overlap between biological and clinical characteristics of the human and canine breast cancer. We attempt to clarify expression profile of the embryonic stem cell (ES) gene signatures in canine breast cancer. Using microarray datasets (GSE22516 and GSE20718), expression of the three major ES gene signatures (modules or gene‐sets), including Myc, ESC‐like, and PRC modules, was primarily analyzed through Gene‐Set Enrichment Analysis (GSEA) method in tumor and healthy datasets. For confirmation of the primary results, an additional 13 ES gene‐sets which were categorized into four groups including ES expressed (ES exp1 and ES exp2), NOS targets (Nanog targets, Oct4 targets, Sox2 targets, NOS targets, and NOS TFs), Polycomb targets (Suz12 targets, Eed targets, H3K27 bound, and PRC2 targets), and Myc targets (Myc targets1, and Myc targets2) were tested in the tumor and healthy datasets. Our results revealed that there is a valuable overlap between canine and human breast cancer ES gene‐sets expression profile, where Myc and ESC‐like modules were up‐regulated and PRC module was down‐regulated in metastatic canine mammary gland tumors. Further analysis of the secondary gene‐sets indicated overexpression of the ES expressed, NOS targets (Nanog targets, Oct4 targets, Sox2 targets, and NOS targets), and Myc targets and underexpression of the Polycomb targets in metastatic canine breast cancer.     

基于改进TF-IDF算法的基因通路富集方法

提出一种综合考虑通路局部和全局信息的基因通路富集分析（GIGSEA）方法。首先利用基因相互作用数据,通过基因在通路的局部重要性和在通路数据库的全局特异性计算基因的影响力;然后将基因影响力和表型相关性值融合在一起,计算通路的富集分数;最后通过置换基因富集出统计学显著的通路。将GIGSEA方法运用于肝细胞癌和结肠直肠癌数据集进行风险通路富集,与基因集富集分析方法相比,GIGSEA方法能富集出一些新的相关通路,并排除无关的通路,提高疾病相关通路的富集效果。     

基于直方图拟合的MRI图像水平集分割

利用混合高斯模型对MRI图像直方图进行分析,将拟合获得的特征参数作为水平集曲线进化的约束条件,对医学图像进行分割。分割中采用的自适应Level Set方法,能够自适应地确定曲线进化方向（扩张或收缩）,而不必在分割之前指定其进化方向,减少了人工干预;同时也克服了传统测地活动轮廓线（GAC）方法对图像梯度信息的过分依赖,以及由于对图像进行大尺度高斯平滑处理造成边缘点移动、定位准确度下降的缺陷。分别对MRI仿真和真实图像进行了实验,MRI仿真实验的分割敏感性、专一性和总体性能指标分别达到了94.72%、97.52%和97.22%。分割结果的定量分析和定性分析表明算法的有效性以及较高的分割准确度。     

基于几何活动轮廓的神经元干细胞序列图像的自动分割

一个快速稳定的分割系统是研究神经元干细胞变化的基础，为完善此系统，针对多连接边缘模糊的细胞分割提取问题，根据曲线进化原理，我们提出了一种基于水平集方法的改进的几何活动轮廓算法。此算法能自动解决图像的拓扑变化，并能获得更加真实的细胞轮廓边缘。将此方法应用于神经元干细胞的序列图像分割，实验结果证明了此算法的有效性与准确性。     

基于改进C-V模型的肿瘤CT图像分割

针对传统C-V模型演化速度慢和不能很好分割灰度不均匀图像的缺点,从两个方面进行了改进。首先采用一个新颖的基于局部梯度的模型,使C-V模型初始轮廓曲线快速移到目标边界附近,大大缩短了演化时间;其次,结合GVF模型从两个方向指向目标边界的特点,为C-V模型的速度方程添加一个自适应速度调节项,使模型收敛于真实边界。通过肝脏肿瘤CT图像的分割,验证该方法是有效的。