基于主成分分析的随机森林视网膜OCT图像分层算法研究

视网膜是层状结构,临床上可以根据视网膜层厚度改变对一些疾病进行预测和诊断.为了快速且准确地分割出视网膜的不同层带,本论文提出一种基于主成分分析的随机森林视网膜光学相干断层扫描技术(optical coherence tomography,OCT)图像分层算法.该方法使用主成分分析(principal component analysis,PCA)法对随机森林采集到的特征进行重采样,保留重采样后权重大的特征信息维度,从而消除特征维度间的关联性和信息冗余.结果表明,总特征维度在29维的情况下,保留前18维度训练速度提高了23.20%,14维度训练速度提高了42.38%,而对图像分割精度方面影响较小,实验表明该方法有效地提高了算法的效率.     

通过SEC-UNet精准分割糖尿病视网膜病变眼底OCT图像脉络膜层

目的 糖尿病视网膜病变（DR）是糖尿病的严重并发症，可导致患者视力下降甚至失明。脉络膜的早期检查在DR诊断中起着至关重要的作用。然而，由于DR患者的光学相干层析成像（OCT）中存在脉络膜和巩膜边界模糊、视网膜病变阴影等问题，导致大多数现有算法无法精准分割脉络膜层。本文目的在于提高DR患者OCT图像中脉络膜层分割的精准度。方法 本文提出了一种结合挤压激励连接（SEC）模块和UNet的网络，简称SEC-UNet，不仅增强Unet的局部细节目标关注能力，且能跳出局部最优来增强整体表达能力。结果 SEC-UNet模型的ROC曲线下面积（AUC）达到0.993 0，优于传统UNet模型和SE-UNet模型。这表明SEC-UNet能够获得准确、完整的脉络膜层分割结果。统计分析脉络膜参数变化发现，与正常眼相比，87.1%的DR患者脉络膜中央凹1 mm内体积增加，这证明了DR很可能导致脉络膜增厚。结论 该技术有望成为一种新的辅助诊断工具，帮助医生研究脉络膜在糖尿病眼病的预防、发病机制和预后中的作用。     

Some Physiological Adjustments Relating to Nutrition in Pregnancy

Information from human and animal studies clearly indicates that a number of adjustments in protein, fat, iron and calcium metabolism occur during pregnancy. Recognition of these adjustments has led to a reappraisal of nutrient requirements during pregnancy. The existence of significant protein storage during pregnancy is called into question; in contrast, the evidence suggests that the major storage component is fat, perhaps in anticipation of the energy costs of lactation. In women who will not lactate, this storage is of no benefit and is an etiologic factor in obesity. From a consideration of iron metabolism in pregnancy, it is concluded that it is better to prevent iron depletion in non-pregnant women, especially that associated with excessive menstrual loss, than to administer iron therapy to pregnant women.     

Anticipatory Adjustments to Being Picked Up in Infancy

Anticipation of the actions of others is often used as a measure of action understanding in infancy. In contrast to studies of action understanding which set infants up as observers of actions directed elsewhere, in the present study we explored anticipatory postural adjustments made by infants to one of the most common adult actions directed to them – picking them up. We observed infant behavioural changes and recorded their postural shifts on a pressure mat in three phases: (i) a prior Chat phase, (ii) from the onset of Approach of the mother’s arms, and (iii) from the onset of Contact. In Study 1, eighteen 3-month-old infants showed systematic global postural changes during Approach and Contact, but not during Chat. There was an increase in specific adjustments of the arms (widening or raising) and legs (stiffening and extending or tucking up) during Approach and a decrease in thrashing/general movements during Contact. Shifts in postural stability were evident immediately after onset of Approach and more slowly after Contact, with no regular shifts during Chat. In Study 2 we followed ten infants at 2, 3 and 4 months of age. Anticipatory behavioural adjustments during Approach were present at all ages, but with greater differentiation from a prior Chat phase only at 3 and 4 months. Global postural shifts were also more phase differentiated in older infants. Moreover, there was significantly greater gaze to the mother’s hands during Approach at 4 months. Early anticipatory adjustments to being picked up suggest that infants’ awareness of actions directed to the self may occur earlier than of those directed elsewhere, and thus enable infants’ active participation in joint actions from early in life.     

Development of a Semi-Automatic Segmentation Method for Retinal OCT Images Tested in Patients with Diabetic Macular Edema

Purpose

To develop EdgeSelect, a semi-automatic method for the segmentation of retinal layers in spectral domain optical coherence tomography images, and to compare the segmentation results with a manual method.

Methods

SD-OCT (Heidelberg Spectralis) scans of 28 eyes (24 patients with diabetic macular edema and 4 normal subjects) were imported into a customized MATLAB application, and were manually segmented by three graders at the layers corresponding to the inner limiting membrane (ILM), the inner segment/ellipsoid interface (ISe), the retinal/retinal pigment epithelium interface (RPE), and the Bruch''s membrane (BM). The scans were then segmented independently by the same graders using EdgeSelect, a semi-automated method allowing the graders to guide/correct the layer segmentation interactively. The inter-grader reproducibility and agreement in locating the layer positions between the manual and EdgeSelect methods were assessed and compared using the Wilcoxon signed rank test.

Results

The inter-grader reproducibility using the EdgeSelect method for retinal layers varied from 0.15 to 1.21 µm, smaller than those using the manual method (3.36–6.43 µm). The Wilcoxon test indicated the EdgeSelect method had significantly better reproducibility than the manual method. The agreement between the manual and EdgeSelect methods in locating retinal layers ranged from 0.08 to 1.32 µm. There were small differences between the two methods in locating the ILM (p = 0.012) and BM layers (p<0.001), but these were statistically indistinguishable in locating the ISe (p = 0.896) and RPE layers (p = 0.771).

Conclusions

The EdgeSelect method resulted in better reproducibility and good agreement with a manual method in a set of eyes of normal subjects and with retinal disease, suggesting that this approach is feasible for OCT image analysis in clinical trials.     

Comparative Physiological Investigation of Peculiarities of Recognition of Complex Visual Images and Ability to Identify Two-Dimensional Images and Three-Dimensional Objects by the Lower and Higher Monkeys

An ability of primates to compare and identify planar images and three-dimensional objects has been studied at solving tasks of the choice on model type. The study is carried out on three species of monkeys of different levels of phylogenetic development: macaques rhesuses Macaca mulata, brown capuchins Cebus apella, and chimpanzee Pan troglodytes. Plaster casts of fruits, household objects, and toys–animals were used for the identification. It is revealed that representatives of the studied monkey species are able to compare and identify three-dimensional objects and their two-dimensional images and to extract identical attributes during solution of visual tasks of different degrees of complexity in variants of the methodical approach choice on model. Under conditions of our experiments, the capuchins, macaques, and chimpanzee performed a direct transfer of the elaborated principle of the solution. There have also been established peculiarities of behavior of different species of the lower and higher monkeys. Macaques rhesuses had difficulties when solving complex combinatory tasks. Capuchins displayed pronounced negative emotions and characteristic patterns of defensive behavior at demonstration and during identification of two- and three-dimensional objects of the toys–animals type. Unlike the lower monkeys, the chimpanzee, when presented with toys–animals, did not change the level of their correct answers and demonstrated pronounced play reactions.     

Using Communication Tools to Facilitate Regulatory Review

Submittals under CERCLA, RCRA, and related environmental regulatory programs often require voluminous data and narrative, supported by multiple appendices of supplemental data. Often lost in the encyclopedic maze of documentation is the key message of the respondent. Also lost is the fact that every document, however data-driven or technical, has a message that should be readable and understandable to both the technical expert and the general reader alike. This is especially true for documents that will undergo public scrutiny at some phase. The regulatory review process can be facilitated if the respondents submitting the documentation present a clear and understandable message that is first understood by the respondents and subsequently by the intended audience. Technical communicators can play an important role in assuring that even a highly complex technical document filled with assumptions, qualifiers, analysis, and quantification retains a clear and understandable message. Simple tools that we may take for granted can be employed throughout the writing process. These tools can range from titles, outlines, and team message slogans to editorial style sheets and review checklists. These tools, used as message reminders, can assist members of a documentation team to stay en pointe, to keep the message in focus while the document is being prepared. Properly applied, these tools translate into a clear message for readers, despite the legitimate and real concerns of sophisticated analysis and scientific inquiry. A clear message, in turn, can expedite the regulatory review process by helping to minimize extraneous comments and excessive requests for clarification from regulatory agencies.     

Procedure to Facilitate Chromosome Counts in Difficult Plant Material

Numerous mitotic plates of contracted and well-spread chromosomes may be obtained from root tips of plants set on melting ice or snow overnight at room temperature. After 1:3 acetic-alcohol fixation for 0.5 to 3 hours the material is mordanted in a mixture of 7 parts of alcohol plus 21/2 parts of 3% ferric ammonium sulfate for 3 hours to overnight. This solution may be used as storage fluid for flower buds. Deep chromosome coloration without precipitates is secured by staining in a few drops of aceto-carmine for 10-15 minutes after which the tissues are softened by heating in aceto-carmine diluted with 3 parts of 45% acetic acid.     

Prefixing with Paradichlorobenzene to Facilitate Chromosome Study

A technic was developed which resulted in preparations containing many mitotic divisions with chromosomes well fixed and stained, rod-shaped, and spread throughout the cell. This technic has given good results with guayule (Parthenium argentatum), Crepis, Allium, Pisum, Lycopersicon, Tradescantia, and other plants. Material is prefixed in a saturated solution of paradichlorobenzene for 1-4 hours, fixed in 65% acetic acid (or other suitable fixative) for 12-24 hours, hydrolyzed in 10% HCl for 10-30 minutes at 60° C, rinsed in water, transferred to a drop of 45% acetic acid on a slide, and smeared and stained in aceto-orcein. The preparation may be made permanent by separating slide and cover glass in 1 part glacial acetic acid to 1 part absolute alcohol, putting them in absolute alcohol, and then recombining them with a drop of euparol.     

Site, Scale and Time-course for Adjustments in Lateral Root Initiation in Wheat Following Changes in C and N Supply

The pattern of lateral root initiation in seminal roots of wheat(Triticum aestivumL. cv. Alexandria) and the location, scaleand time-course for adjustments in initiation were studied afterchanges in C and N supply. Macroscopically visible primordiaappeared in a non-acropetal sequence with the frequency (numberper unit length) increasing with distance behind the main rootapex to a maximum at 40–50 mm behind the root tip. Pruningthe root system to a single seminal axis increased the primordiafrequency by 23% within 15 h. After longer periods, the effectof root-pruning was greater. The enhanced primordia frequencywas first observed in tissue located 0–10 mm behind theapex at the start of treatment. Feeding glucose (50 mM) alsoincreased primordia frequency within 15 h, but to a greaterextent, and here additional primordia were initiated in tissuelocated 0–10and10–20 mm behind the apex at the startof treatment. Withdrawing NO₃^-from one part of a split-rootsystem, whilst maintaining the supply to the other, reducedprimordia frequency in the non-fed roots and, in some cases,a compensatory increase in the NO₃^--fed roots was observed.The location and scale of the adjustments were similar to thosefound with root-pruning and glucose-feeding, but were slightlyslower to appear. In spite of some differences in detail, therewas a broad similarity in site, scale and time-course for adjustmentsin lateral root initiation with these treatments, which is consistentwith the operation of a common mechanism. Whenever an increasein primordia frequency was observed, it was associated withan increase in the ethanol-soluble sugar content of the tissue.However, the reduction in frequency in NO₃^--deprived roots wasalso accompanied by an increase in sugar content. There wasno consistent relationship between total N content of the tissueand primordia frequency, but there was between primordia frequencyand the rate of net NO₃^-uptake. The possible mechanisms controllinglateral root initiation are discussed. Compensatory growth; correlative growth; glucose; initiation; lateral root; nitrate; primordium; split-root; Triticum aestivum; wheat     

Chromosomal Fusions Facilitate Adaptation to Divergent Environments in Threespine Stickleback

Chromosomal fusions are hypothesized to facilitate adaptation to divergent environments, both by bringing together previously unlinked adaptive alleles and by creating regions of low recombination that facilitate the linkage of adaptive alleles; but, there is little empirical evidence to support this hypothesis. Here, we address this knowledge gap by studying threespine stickleback (Gasterosteus aculeatus), in which ancestral marine fish have repeatedly adapted to freshwater across the northern hemisphere. By comparing the threespine and ninespine stickleback (Pungitius pungitius) genomes to a de novo assembly of the fourspine stickleback (Apeltes quadracus) and an outgroup species, we find two chromosomal fusion events involving the same chromosomes have occurred independently in the threespine and ninespine stickleback lineages. On the fused chromosomes in threespine stickleback, we find an enrichment of quantitative trait loci underlying traits that contribute to marine versus freshwater adaptation. By comparing whole-genome sequences of freshwater and marine threespine stickleback populations, we also find an enrichment of regions under divergent selection on these two fused chromosomes. There is elevated genetic diversity within regions under selection in the freshwater population, consistent with a simulation study showing that gene flow can increase diversity in genomic regions associated with local adaptation and our demographic models showing gene flow between the marine and freshwater populations. Integrating our results with previous studies, we propose that these fusions created regions of low recombination that enabled the formation of adaptative clusters, thereby facilitating freshwater adaptation in the face of recurrent gene flow between marine and freshwater threespine sticklebacks.     

Policies and Programs to Facilitate Access to Targeted Cancer Therapies in Thailand

Background

Increasing access to clinically beneficial targeted cancer medicines is a challenge in every country due to their high cost. We describe the interplay of innovative policies and programs involving multiple stakeholders to facilitate access to these medicines in Thailand, as well as the utilization of selected targeted therapies over time.

Methods

We selected two medicines on the 2013 Thai national list of essential medicines (NLEM) [letrozole and imatinib] and three unlisted medicines for the same indications [trastuzumab, nilotinib and dasatinib]. We created timelines of access policies and programs for these products based on scientific and grey literature. Using IMS Health sales data, we described the trajectories of sales volumes of the study medicines between January 2001 and December 2012. We compared estimated average numbers of patients treated before and after the implementation of policies and programs for each product.

Results

Different stakeholders implemented multiple interventions to increase access to the study medicines for different patient populations. During 2007–2009, the Thai Government created a special NLEM category with different coverage requirements for payers and issued compulsory licenses; payers negotiated prices with manufacturers and engaged in pooled procurement; pharmaceutical companies expanded patient assistance programs and lowered prices in different ways. Compared to before the interventions, estimated numbers of patients treated with each medicine increased significantly afterwards: for letrozole from 645 (95% CI 366–923) to 3683 (95% CI 2,748–4,618); for imatinib from 103 (95% CI 72–174) to 350 (95% CI 307–398); and for trastuzumab from 68 (95% CI 45–118) to 412 (95% CI 344–563).

Conclusions

Government, payers, and manufacturers implemented multi-pronged approaches to facilitate access to targeted cancer therapies for the Thai population, which differed by medicine. Routine monitoring is needed to assess clinical and economic impacts of these strategies in the health system.     

Recovering Root System Traits Using Image Analysis Exemplified by Two-Dimensional Neutron Radiography Images of Lupine

Root system traits are important in view of current challenges such as sustainable crop production with reduced fertilizer input or in resource-limited environments. We present a novel approach for recovering root architectural parameters based on image-analysis techniques. It is based on a graph representation of the segmented and skeletonized image of the root system, where individual roots are tracked in a fully automated way. Using a dynamic root architecture model for deciding whether a specific path in the graph is likely to represent a root helps to distinguish root overlaps from branches and favors the analysis of root development over a sequence of images. After the root tracking step, global traits such as topological characteristics as well as root architectural parameters are computed. Analysis of neutron radiographic root system images of lupine (Lupinus albus) grown in mesocosms filled with sandy soil results in a set of root architectural parameters. They are used to simulate the dynamic development of the root system and to compute the corresponding root length densities in the mesocosm. The graph representation of the root system provides global information about connectivity inside the graph. The underlying root growth model helps to determine which path inside the graph is most likely for a given root. This facilitates the systematic investigation of root architectural traits, in particular with respect to the parameterization of dynamic root architecture models.Crucial factors for plant development are light quantity and quality as well as water and nutrient availability in soils. Regarding water and nutrient uptake, root architecture is the main aspect of plant productivity (Lynch, 2007; Smith and De Smet, 2012) and needs to be accurately considered when describing root processes. Currently, understanding the impact of roots and rhizosphere traits on plant resource efficiency is of highest relevance (Hinsinger et al., 2011). Development in this area will increase food security by enabling more sustainable production with reduced fertilizer input by improving cropping systems and cultivars for resource-limited environments (de Dorlodot et al., 2007).Root architectural development includes architectural, morphological, anatomical, as well as physiological traits. For the systematic investigation of such complex biological systems, mathematical modeling is inevitable (Roose and Schnepf, 2008). Ideally, experiments and theoretical models are developed to mutually support each other. In this way, models are created that include state-of-the-art knowledge and have significant parameters. There are various root architectural models incorporating a multitude of processes (Dunbabin et al., 2013) that are originally based on Pagès et al. (1989) and Diggle (1988). Generally, the parameterization of such models is difficult and demands elaborate experimental effort. In this work, we present a novel approach for recovering root system parameters based on image-analysis techniques. In this way, we simplify the systematic investigation of root architectural traits, in particular with respect to the parameterization of root system models.Imaging techniques for the visualization of soil-grown root systems in two and three dimensions include x-ray computed tomography (Heeraman et al., 1997; Tracy et al., 2010; Mooney et al., 2012), neutron radiography (NR; Oswald et al., 2008), and magnetic resonance imaging (Pohlmeier et al., 2008). NR is one of the most suitable techniques to investigate roots grown in soil, because it allows a high throughput, provides a strong contrast between roots and soil, and therefore requires little effort for image processing. A major advantage of NR as well as magnetic resonance imaging is the possibility to monitor water distribution and roots simultaneously (Menon et al., 2007; Oswald et al., 2008; Moradi et al., 2009; Carminati et al., 2010; Stingaciu et al., 2013). This is especially useful as water is a crucial factor ruling root allocation in soil (Hodge, 2010).Images of root architecture contain a huge amount of information, and image analysis helps to recover parameters describing certain root architectural and morphological traits. The majority of imaging systems for root systems are designed for two-dimensional images, such as RootReader2D (Clark et al., 2013), GiA Roots (Galkovskyi et al., 2012), SmartRoot (Lobet et al., 2011), EZ-Rhizo (Armengaud et al., 2009), and Growscreen (Nagel et al., 2012). See also Le Bot et al. (2010) for a review of available software. A starting point for image analysis is commonly a grayscale image of a root system. The first step is to create a binary image by segmentation. Further steps include skeletonization, root tracking, and data analysis. The most common segmentation method is some form of thresholding (e.g. RootReader2D, GiA Roots, SmartRoot, EZ-Rhizo; Stingaciu et al. [2013]). Other methods include the livewire algorithm (Basu and Pal, 2012) or the levelset method (RootTrak; Mairhofer et al., 2012) that determine the borders of each root. The creation of a root system skeleton is either done manually (e.g. DART; Le Bot et al., 2010) or based on morphological operators such as thinning and closing (GiA Roots, RootReader2D, EZ-Rhizo), sometimes with options for the user to correct skeleton points (EZ-Rhizo). The root tracking step can be performed manually (DART) or based on creating a graph representation of the root system combined with Dijkstra’s algorithm, a search algorithm that finds the shortest path between two nodes inside a graph (RootReader2D; Stingaciu et al. [2013]). Furthermore, algorithms can operate on the skeleton (EZ-Rhizo) or directly on the image source (SmartRoot). In SmartRoot, the user selects a root in the original image with a mouse click and then a skeletonization algorithm determines the skeleton of the selected root. The output of all root tracking algorithms is a data structure of a set of roots that stores information such as connectivity between roots and their position in space.Global traits of the root system are obtained directly from the segmented image or the skeleton (GiA Roots, RooTrak). Global traits include convex hull, network depth, network length distribution, maximum number of roots, maximum width of the root system, network length, and specific root length. Furthermore, the data structure from a root tracking procedure is used to obtain individual, local root parameters (DART, RootReader2D, EZ-Rhizo, SmartRoot). The latter are able to obtain root architectural parameters that can be used for model parameterization. An additional aspect is the dealing with dynamic data (i.e. images of the same root system taken at several times). Analysis of such sequences may lead to better insight into the development of the root system (e.g. DART, SmartRoot) or even reveal growth zones and their local growth velocities (Basu and Pal, 2012).Analysis software for two-dimensional images of soil-grown root systems currently work in a semiautomated way with respect to tracing individual roots. This requires considerable user input for larger root systems. We present a new, fully automated approach for recovering root architectural parameters from two-dimensional images of root systems. The software Root System Analyzer is, to our knowledge, the first algorithm for two-dimensional analysis of soil-grown root systems that features fully automated root tracking. Only primary roots have to be initiated manually by the user. The user is also free to initiate any laterals, but this is not mandatory. Further growth of primary roots and laterals is then tracked in a fully automated way. In addition, there is a user interface that allows for manual correction of individual roots if required. In this work, we do not go into the details about the segmentation step but focus on the root tracking step and the parameterization of a root system model (Leitner et al., 2010). The described algorithm starts with a sequence of segmented two-dimensional images showing the dynamic development of a root system. For each image, morphological operators are used for skeletonization. Based on this, a graph representation of the root system is created. A dynamic root architecture model helps to determine which edges of the graph belong to an individual root. The algorithm elongates each root at the root tip and simulates growth confined within the already existing graph representation. The increment of root elongation is calculated assuming constant growth. For each root, the algorithm finds all possible paths and elongates the root in the direction of the optimal path. In this way, each edge of the graph is assigned to one or more coherent roots. The algorithm considers the fact that new branches can only emerge after the apical zone has developed, which helps in the decision of whether the root is branching or two roots are crossing or overlapping. Image sequences of root systems are handled in such a way that the previous image is used as a starting point for the current image. This is helpful in the analysis of complex root systems as well as for retrieving dynamic parameters such as elongation rates. The algorithm is implemented in a set of Matlab m-files, which makes the code flexible so that it can easily be adjusted to specific experimental setups or mathematical models.We exemplify the approach with two-dimensional neutron radiography images of lupine (Lupinus albus) root systems grown in mesocosms filled with a sandy soil. Furthermore, we compare our approach with the approaches of SmartRoot and RootReader2D and demonstrate how our approach can be used to analyze large root systems.     

Recovery of the Two-Dimensional Ion Distribution Function in a Magnetic Mirror from Measurements of Collective Thomson Scattering Spectra

Plasma Physics Reports - A method is proposed for tomography of the distribution function of energetic ions that are adiabatically trapped in an open magnetic trap, according to the diagnostic data...