CT图像肺及肺病变区域分割方法综述

计算机断层扫描（computed tomography,CT）技术能为新冠肺炎（corona virus disease 2019,COVID-19）和肺癌等肺部疾病的诊断与治疗提供更全面的信息,但是由于肺部疾病的类型多样且复杂,使得对肺CT图像进行高质量的肺病变区域分割成为计算机辅助诊断的重难点问题。为了对肺CT图像的肺及肺病变区域分割方法的现状进行全面研究,本文综述了近年国内外发表的相关文献：对基于区域和活动轮廓的肺CT图像传统分割方法的优缺点进行比较与总结,传统的肺CT图像分割方法因其实现原理简单且分割速度快等优点,早期使用较多,但其存在分割精度不高的缺点,目前仍有不少基于传统方法的改进策略;重点分析了基于卷积神经网络（convolutional neural network,CNN）、全卷积网络（fully convolutional network,FCN）、U-Net和生成对抗网络（generative adversarial network,GAN）的肺CT图像分割网络结构改进模型的研究进展,基于深度学习的分割方法具有分割精度高、迁移学习能力强和鲁棒性高等优点,特别是在辅助诊断COVID-19病例时,基于深度学习方法的性能明显优于基于传统方法的性能;介绍肺及肺病变区域分割的常用数据集和评价指标,在解决如COVID-19数据样本量少等问题时,使用GAN以合成高质量的对抗性图像用以扩充数据集,从而增加训练样本的数量和多样性;讨论了肺CT图像的肺及肺病变区域的高精度分割策略的研究趋势、现有挑战和未来的研究方向。     

Genetic parameters for hoof health traits estimated with linear and threshold models using alternative cohorts

A national genetic evaluation program for hoof health could be achieved by using hoof lesion data collected directly by hoof trimmers. However, not all cows in the herds during the trimming period are always presented to the hoof trimmer. This preselection process may not be completely random, leading to erroneous estimations of the prevalence of hoof lesions in the herd and inaccuracies in the genetic evaluation. The main objective of this study was to estimate genetic parameters for individual hoof lesions in Canadian Holsteins by using an alternative cohort to consider all cows in the herd during the period of the hoof trimming sessions, including those that were not examined by the trimmer over the entire lactation. A second objective was to compare the estimated heritabilities and breeding values for resistance to hoof lesions obtained with threshold and linear models. Data were recorded by 23 hoof trimmers serving 521 herds located in Alberta, British Columbia, and Ontario. A total of 73,559 hoof-trimming records from 53,654 cows were collected between 2009 and 2012. Hoof lesions included in the analysis were digital dermatitis, interdigital dermatitis, interdigital hyperplasia, sole hemorrhage, sole ulcer, toe ulcer, and white line disease. All variables were analyzed as binary traits, as the presence or the absence of the lesions, using a threshold and a linear animal model. Two different cohorts were created: Cohort 1, which included only cows presented to hoof trimmers, and Cohort 2, which included all cows present in the herd at the time of hoof trimmer visit. Using a threshold model, heritabilities on the observed scale ranged from 0.01 to 0.08 for Cohort 1 and from 0.01 to 0.06 for Cohort 2. Heritabilities estimated with the linear model ranged from 0.01 to 0.07 for Cohort 1 and from 0.01 to 0.05 for Cohort 2. Despite a low heritability, the distribution of the sire breeding values showed large and exploitable variation among sires. Higher breeding values for hoof lesion resistance corresponded to sires with a higher prevalence of healthy daughters. The rank correlations between estimated breeding values ranged from 0.96 to 0.99 when predicted using either one of the 2 cohorts and from 0.94 to 0.99 when predicted using either a threshold or a linear model.     

A deep supervised approach for ischemic lesion segmentation from multimodal MRI using Fully Convolutional Network

The principle restorative step in the treatment of ischemic stroke depends on how fast the lesion is delineated from the Magnetic Resonance Imaging (MRI) images. This will serve as a vital aid to estimate the extent of damage caused to the brain cells. However, manual delineation of the lesion is time-consuming and it is subjected to intra-observer and inter-observer variability. Most of the existing methods for ischemic lesion segmentation rely on extracting handcrafted features followed by application of a machine learning algorithm. Identifying such features demand multi-domain expertise in Neuro-radiology as well as Image processing. This can be accomplished by learning the features automatically using Convolutional Neural Network (CNN). To perform segmentation, the spatial arrangement of pixel needs to be preserved in addition to learning local features of an image. Hence, a deep supervised Fully Convolutional Network (FCN) is presented in this work to segment the ischemic lesion. The highlight of this research is the application of Leaky Rectified Linear Unit activation in the last two layers of the network for a precise reconstruction of the ischemic lesion. By doing so, the network was able to learn additional features which are not considered in the existing U-Net architecture. Also, an extensive analysis was conducted in this research to select optimal hyper-parameters for training the FCN. A mean segmentation accuracy of 0.70 has been achieved from the experiments conducted on ISLES 2015 dataset. Experimental observations show that our proposed FCN method is 10% better than the existing works in terms of Dice Coefficient.     

积极降压对脑小血管病防治策略的影响

2017年,美国心脏病学会和美国心脏协会提出将高血压诊断标准降为130/80 mm Hg (1 mm Hg=0. 133 kPa),这与其他指南有所不同。降压治疗的核心目标在于靶器官的保护,近年来降压治疗与脑小血管病预防的相关问题日益受到临床重视。目前研究结果表明,降压治疗对腔隙性卒中患者的二级预防以及白质病变进展的预防可能有一定积极意义,但确切的降压目标值并未确定。血压与临床结局可能呈现J型关系,血压过低或过高可能均有害,而取得最大获益的降压目标有待进一步探索。     

Short communication: Genetic characterization of digital cushion thickness

Dairy cow lameness is a serious animal welfare issue. It is also a significant cause of economic losses, reducing reproductive efficiency and milk production and increasing culling rates. The digital cushion is a complex structure composed mostly of adipose tissue located underneath the distal phalanx and has recently been phenotypically associated with incidence of claw horn disruption lesions (CHDL); namely, sole ulcers and white line disease. The objective of this study was to characterize digital cushion thickness genetically and to investigate its association with body condition score (BCS), locomotion score (LOCO), CHDL, and milk production. Data were collected from 1 large closely monitored commercial dairy farm located in upstate New York; 923 dairy cows were used. Before trimming, the following data were collected by a member of the research team: BCS, cow height measurement, and LOCO. Presence or not of CHDL (sole ulcer or white line disease, or both) was recorded at trimming. Immediately after the cows were hoof trimmed, they underwent digital sonographic B-mode examination for the measurement of digital cushion thickness. Factors such as parity number, stage of lactation, calving date, mature-equivalent 305-d milk yield (ME305MY), and pedigree information were obtained from the farm’s dairy management software (DairyCOMP 305; Valley Agricultural Software, Tulare, CA). Univariate animal models were used to obtain variance component estimations for each studied trait (CHDL, BCS, digital cushion thickness average, LOCO, height, and ME305MY) and a 6-variate analysis was conducted to estimate the genetic, residual, and phenotypic correlations between the studied traits. The heritability estimate of DCTA was 0.33 ± 0.09, whereas a statistically significant genetic correlation was estimated between DCTA and CHDL (−0.60 ± 0.29). Of the other genetic correlations, significant estimates were derived for BCS with LOCO (−0.49 ± 0.19) and ME305MY (−0.48 ± 0.20). Digital cushion thickness is moderately heritable and genetically strongly correlated with CHDL.     

脑白质损伤分割方法比较和策略改进

当前对白质损伤分割研究的缺陷就是不能真正用于实时分割任务,其主要原因是所用特征的表现性能较低.针对此问题,提出了新的数据处理方法和损伤分割特征.方法通过划分数据减少处理数据量的同时采用分治策略,区分处理白质灰质区域,从而改善了整体白质损伤分割的性能.在ACCORD-MIND MRI数据集上,使用SVM分类器将该新特征和传统特征的分割性能作了比较.实验结果表明,该特征在白质损伤分割中优于传统的特征,并能获得较好的分割结果和较快的分割速度.     

Measures of weight distribution of dairy cows to detect lameness and the presence of hoof lesions

There is increasing interest in automated methods of detecting lame cows. Hoof lesion data and measures of weight distribution from 61 lactating cows were examined in this study. Lame cows were identified with different numerical rating scores (NRS) used as thresholds (NRS >3 and NRS ≥3.5) for lameness. The ratio of weight applied to a pair of legs (LWR) when the cow was standing was calculated using a special weigh scale, and the cows were gait scored using a 1 to 5 NRS. Hoof lesions were scored and the cows placed into 1 of 4 mutually exclusive categories of hoof lesion: a) no lesions, b) moderate or severe hemorrhages, c) digital dermatitis, and d) sole ulcers. Regression analysis and receiver operating characteristic (ROC) curves were used to analyze the relation between hoof lesions and LWR. A clear relationship was found between NRS and LWR for the cows with sole ulcers (R² = 0.79). The LWR could differentiate cows with sole ulcers from sound cows with no hoof lesions [area under the curve (AUC) = 0.87] and lame cows from nonlame cows with lameness thresholds NRS >3 (AUC = 0.71) and NRS ≥3.5 (AUC = 0.88). There was no relationship between LWR and NRS for cows with digital dermatitis. Measurement of how cows distribute their weight when standing holds promise as a method of automated detection of lameness.