基于层间先验知识从脑MRI图像中自动提取脑组织

目的从脑MRI图像中提取脑组织,解决边缘模糊时脑和非脑组织难以分离的问题。方法首先利用各向异性扩散滤波的方法对脑MRI图像进行去噪处理;然后利用形态学的方法对初始脑MRI图像进行脑组织提取,在此分割结果的基础上,利用相邻层脑形态差异较小的特点,实现结构元素的自适应选取,完成从脑MRI图像中逐层准确、自动提取脑组织。结果采用不同来源的数据对算法性能进行了测试,结果优于经典背散射电子成像(BSE)方法的分割结果。结论利用层间先验知识有利于实现边缘模糊的脑组织自动准确提取,且适用性较强。     

基于双通道三维密集连接网络的脑胶质瘤核磁共振成像分割算法研究

针对脑胶质瘤形状、位置及大小的不一致性,本文提出了一种基于双通道三维密集连接网络的脑胶质瘤核磁共振成像(MRI)自动分割算法。该算法基于三维卷积神经网络,在两个通道采用不同大小卷积核,从而在不同尺度感受野下提取多尺度特征,并构造各自的密集连接块进行特征学习与传递,通过特征结联后输入到分类层进行目标体素分类,最终实现脑胶质瘤的自动分割。为了验证本文算法的实用性,本文采用公开的脑肿瘤分割挑战赛数据集对网络进行训练与验证,并将得到的结果与其他脑胶质瘤分割方法比较。实验结果表明,本文所提出的算法能够更准确地分割出不同的肿瘤病变区域,在临床脑肿瘤疾病诊断中具有一定的应用价值。     

基于边界跟踪算法的脑磁共振图像分割

目的：脑磁共振图像的自动分割是近几年研究的一大热点,本文在通过分析比较当前各种图像分割算法后,介绍了一种基于边界跟踪的脑磁共振图像（MRI）分割算法,在MRI中提取出脑组织部分。方法：应用迭代法对脑磁共振图像进行二值化处理;扫描二值化图像,根据脑组织的形态,确定一点作为脑组织边界的起点;根据边界点的像素特征,对传统的边界跟踪算法进行改进,计算出MRI脑组织的边界,最后应用区域生长法在原始MRI中提取脑组织图像,实现MRI分割。结果：实验结果表明,改进后的边界跟踪算法在提取脑组织边界时,细节处理能力强,不易陷入死循环,具有较高的运算速度。提取的真实脑磁共振图像的脑组织区域,能满足临床的实际需要。结论：对图像二值化处理,对图像有微弱的损害,但是极大地提高了计算速度。与传统的边界跟踪算法相比,改进后的边界跟踪算法分割效率高,更易实现MRI的自动分割。获得的边界曲线在细节上更接近于脑组织的实际边界。     

深度卷积神经网络在放射治疗计划图像分割中的应用

目的:结合全卷积神经网络(Fully Convolutional Network,FCN)和多孔卷积(Atrous Convolution,AC)的深度学习方法,实现放射治疗计划图像的组织器官自动勾画。方法:选取122套已经由放疗医师勾画好正常器官结构轮廓的胸部患者CT图像,以其中71套图像(8 532张轴向切层图像)作为训练集,31套图像(5 559张轴向切层图像)作为验证集,20套图像(3 589张轴向切层图像)作为测试集。选取5种公开的FCN网络模型,并结合FCN和AC算法形成3种改进的深度卷积神经网络,即带孔全卷积神经网络(Dilation Fully Convolutional Network,D-FCN)。分别以训练集图像对上述8种网络进行调优训练,使用验证集图像在训练过程中对8种神经网络进行器官自动识别勾画验证,以获取各网络的最佳分割模型,最后使用测试集图像对充分训练后获取的最佳分割模型进行勾画测试,比较自动勾画与医师勾画的相似度系数(Dice)评价各模型的图像分割能力。结果:使用训练图像集进行充分调优训练后,实验的各个神经网络均表现出较好的自动图像分割能力,其中改进的D-FCN 4s网络模型在测试实验中具有最佳的自动分割效果,其全局Dice为94.38%,左肺、右肺、心包、气管和食道等单个结构自动勾画的Dice分别为96.49%、96.75%、86.27%、61.51%和65.63%。结论:提出了一种改进型全卷积神经网络D-FCN,实验测试表明该网络模型可以有效地提高胸部放疗计划图像的自动分割精度,并可同时进行多目标的自动分割。     

基于三重注意力的脑肿瘤图像分割网络

脑肿瘤图像分割问题是脑肿瘤临床诊断和治疗脑肿瘤疾病计算机辅助诊断的基础.针对脑肿瘤MRI图像分割网络深度过深和局部与全局特征信息联系匮乏导致图像分割精度降低等问题,提出一种基于三重注意力的脑肿瘤图像分割网络.首先,借鉴残差结构,将原始图像分割网络结构的编码层和解码层中的卷积模块替换为深度残差模块,解决网络加深带来的梯度...     

基于有限训练样本的融合网络模型用于盆腔危及器官自动分割的研究

将深度学习应用到医学影像中危及器官自动分割领域时,为解决训练样本不足时三维卷积神经网络优化出现的退化、梯度消失等问题,本研究将Dense Net与V-Net两个网络模型进行融合,开发一种用于三维计算机断层扫描(CT)图像自动分割的Dense V-Network算法,勾画女性盆腔危及器官。采用戴斯相似性系数(DSC)、豪斯多夫距离(HD)、杰卡德距离(JD)三个参数来定量评估分割效果。结果显示膀胱、小肠、直肠、股骨头和脊髓自动分割的DSC值均在0.87以上(平均值是0.9);JD值均在2.3以内(平均值是0.18);除小肠外,HD值均在0.9 cm以内(平均值是0.62 cm)。经验证,Dense V-Network网络可精准地勾画盆腔危及器官。     

基于3D卷积网络和多模态MRI的脑胶质瘤自动分割

目的:基于三维（3D）卷积神经网络和多模态MRI实现脑胶质瘤的自动分割。方法:首先对来自BraTS2020公共数据集的369例脑胶质瘤的4个模态MRI数据进行3D剪裁、重采样、去伪影、归一化的预处理。其次将MRI数据和脑胶质瘤标注信息输入到基于U-net的3D卷积神经网络模型进行训练和测试。利用相似性系数、召回率和精确率评价整体肿瘤区域、核心肿瘤区和增强肿瘤区的分割结果。结果:在74例测试数据集上,整体肿瘤区域、核心肿瘤区域和增强肿瘤区域的相似系数平均值分别为0.88、0.77和0.73,中位值分别为0.90、0.84和0.81,召回率平均值分别为0.88、0.78和0.78,中位值分别为0.90、0.84和0.84,精确率平均值分别为0.89、0.83和0.75,中位值分别为0.91、0.89和0.79。结论:基于U-net的3D卷积神经网络在多模态MRI数据集上获得了较好的分割结果,显示其在脑胶质瘤自动分割方面的潜力,可为临床诊断分级和治疗策略选择提供参考。     

基于3D U-Net 实现人体耳软骨MRI图像的解剖结构分割

自体肋软骨雕刻法是目前治疗先天性小儿畸形的临床标准疗法,而耳软骨组织工程和3D生物打印是有前景的治疗方案。可是,这些治疗方案的核心—(复合物)支架构造缺乏基于医学图像的耳软骨自动分割方法。基于3D U-Net提出改进的网络模型,能够实现MRI图像的人体耳软骨解剖结构的自动分割。该网络模型结合残差结构和多尺度融合等设计,在减少网络参数量的同时实现12个耳软骨解剖结构的精确分割。首先,使用超短回波时间(UTE)序列采集40名志愿者单侧外耳的MRI图像;然后,对所采集的图像进行预处理、耳软骨和多解剖结构手动标注;接下来,划分数据集训练改进的3D U-Net模型,其中32例数据作为训练集、4例为验证集、4例为测试集;最后,使用三维全连接条件随机场对网络输出结果进行后处理。模型经过10折交叉验证后,耳软骨12个解剖结构的自动分割结果的平均Dice相似度系数(DSC)和平均95%豪斯多夫距离(HD95)分别为0.818和1.917,相比于使用基础的3D U-Net模型,DSC指标分别提高6.0%,HD95指标降低了3.186,其中耳软骨关键结构耳轮和对耳轮的DSC指标达到了0.907和0.901。实验结果表明,所提出的深度学习方法与专家手动标注两者之间的结果非常接近。在临床应用中,根据患者健侧UTE核磁图像,本研究提出的方法既可以为现有自体肋软骨雕刻法快速、自动生成三维个性化雕刻模板,也可以为组织工程或者3D生物打印技术构建耳软骨复合物支架提供高质量的可打印模型。     

基于DenseMedic网络的脑皮层下结构的语义分割

脑皮层下结构分割问题是神经科及其他相关疾病计算机辅助诊断和治疗的基础。通过分割和分析核磁共振图像中的脑结构,可以对自闭症谱系障碍、脑卒中、脑肿瘤等疾病进行早期诊断和治疗。为解决精准脑结构分割的问题,基于深度学习基本理论,提出一种DenseMedic网络的核磁共振图像脑皮层下结构的分割算法。首先,OreoDown方法通过较早地增大卷积核的步长增大特征感受野的增长速度,并使用不变尺寸的卷积层夹心式地恢复网络深度,使速度的增加带来有效的感受野增加;其次,DenseMedic使用DenseNet的思想实例化OreoDown框架,通过密集连接的特征提取操作来获取多尺度的上下文信息;最后,在各层中使用混合空洞卷积进一步扩大感受野,解决特征感知过于粗糙的问题。采用Dice相似度系数(DSC)、交并比(IoU)、95% Hausdorff表面距离(HSD95)和平均表面距离(ASD) 4个指标,评价神经网络的分割性能。在公开的IBSR数据集的18例图像上进行实验,算法的4个指标分别达到89.2%、80.7%、1.982和0.882;在公开的MBBrainS18数据集的7例图像上的实验显示,算法的4个指标分别达到88.7%、79.8%、1.249和0.570。实验表明,所提出的算法使脑结构的分割结果与真实结构在区域上有更多的重叠, 在轮廓上更加相似,可以更好地完成各个脑皮层下结构的分割。在临床应用中,对脑皮层下结构的精准分割将有助于准确测量相关疾病诊断的关键指标,并实现快速的计算机辅助治疗。     

基于深度学习的轴突三维图像分割与重构

目的:设计适用于神经轴突荧光显微图像三维分割任务的深度神经网络(DNN),提高神经轴突自动重构的准确性。方法:将三维图像分割任务转化为在3个正交投影方向上的二维分割任务,在开源网络DeepMACT的基础上,利用多通道输入降低压缩维度上的信息损失,利用神经轴突骨架加权的损失函数训练以强调神经骨架连贯性,提出SWCUnet。采用转盘共聚焦荧光显微成像采集的小鼠大脑稀疏标记神经元神经轴突部分49个图块作为数据集,以人工标注的形态重构结果作为金标准训练网络。对网络模型输出的二维分割图像进行三维复原,并以三维分割结果输入MOST算法进行自动重构。结果:SWCUnet (32通道,骨架权重5)三维分割F1-score达到0.662,较DeepMACT提升0.132。基于三维分割结果的自动重构F1-score达到0.80,比基于原图的自动重构提升0.24。结论:SWCUnet可以较好地提取高分辨荧光显微图像中的神经轴突特征,输出的三维分割结果实现了大幅提升神经轴突骨架形态的自动重构准确率的目标,为小鼠大脑稀疏标记神经元全脑成像数据的大规模自动化重构提供了一种新工具。     

The serial reaction time task revisited: a study on motor sequence learning with an arm-reaching task

With a series of novel arm-reaching tasks, we have shown that visuomotor sequence learning encompasses the acquisition of the order of sequence elements, and the ability to combine them in a single, skilled behavior. The first component, which is mostly declarative, is reflected by changes in movement onset time (OT); the second, which occurs without subject’s awareness, is measured by changes in kinematic variables, including movement time (MT). Key-press-based serial reaction time tasks (SRTT) have been used to investigate sequence learning and results interpreted as indicative of the implicit acquisition of the sequence order. One limitation to SRT studies, however, is that only one measure is used, the response time, the sum of OT and MT: this makes interpretation of which component is learnt difficult and disambiguation of implicit and explicit processes problematic. Here, we used an arm-reaching version of SRTT to propose a novel interpretation of such results. The pattern of response time changes we obtained was similar to the key-press-based tasks. However, there were significant differences between OT and MT, suggesting that both partial learning of the sequence order and skill improvement took place. Further analyses indicated that the learning of the sequence order might not occur without subjects’ awareness. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Motor sequence consolidation: constrained by critical time windows or competing components

Skill improvements may develop between practice sessions during memory consolidation. Skill enhancement within an egocentric coordinate frame develops over wake, whereas skill enhancement in an allocentric coordinate frame develops over a night of sleep. We tested whether both types of improvement could develop over two different 24-h intervals: 8 am to 8 am or from 8 pm to 8 pm. We found that for each 24 h interval, only one type of skill improvement was seen. Despite passing through wake and a night of sleep participants only showed skill improvements commensurate with either a night of sleep or a day awake. The nature of the off-line skill enhancement was determined by when consolidation occurred within the normal sleep–wake cycle. We conclude that motor sequence consolidation is constrained either by having critical time windows or by a competitive interaction in which improvements within one co-ordinate frame actively block improvements from developing in the alternative co-ordinate frame.     

Convergent results in eyeblink conditioning and contingency learning in humans: Addition of a common cue does not affect feature-negative discriminations

Previous human discrimination learning experiments with eyeblink conditioning showed that an increase in the similarity between the to-be-discriminated stimuli had no effect on the rate of learning. This result was at variance with data from other experiments which had used different paradigms and different stimulus materials. We therefore compared human discrimination learning in eyeblink conditioning and contingency learning using carefully matched procedures. Participants learned two feature-negative discriminations, A+/AB− and CD+/CDE−. Convergent results were obtained in both paradigms. Adding a common cue did not affect response differentiation, i.e. the A+/AB− discrimination and the CD+/CDE− discriminations were equivalent. These results support the notion that learning in both paradigms is based on the same principles. However, the overall pattern of results cannot be easily accommodated within associative learning theories based on the Rescorla-Wagner Model or on Pearce's Configural Model. The application of these models to current and previous data is discussed.     

Engaging patients in health care: an empirical study of the role of engagement on attitudes and action

Objective

Identify the role of engaging people affected by cancer in service development in influencing healthcare professionals and service-users’ attitudes toward, and enactment of, engagement.

Methods

Focus group discussions with healthcare professionals and people affected by lung cancer, prior to and following an intervention where lung cancer teams were supported to engage with patients and family members.

Results

Staff and people affected by cancer who participated displayed more positive attitudes toward involvement than those who did not participate.

Conclusion

Progressing the involvement agenda requires the use of supported, small scale, projects where staff and patients/family members can develop their skills and knowledge of involvement.

Practice implications

Doing patient engagement work is a vital step toward changing attitudes and actions toward the involvement agenda.     

The problem of reversals in assessing implicit sequence learning with serial reaction time tasks

We report two experiments in which implicit learning is demonstrated within a short session of practice, in the absence of explicit knowledge of what is learned. In Experiment 1, we replicate the experiments by Curran (Psychol Res 60:24-41, 1997a; J Cogn Neurosci 9(4):522-533, 1997b) and highlight the importance of avoiding a random sequence as comparison to the training sequence, due to the higher proportion of reversal trials included in the random one, which leads to an artifactual measure of learning. In Experiment 2 we present a procedure in which two structurally analogous sequences are used both as training and control sequences, thus controlling for any factor different from learning. The results show that implicit learning is obtained within a short session of practice, and in the absence of any explicit knowledge as assessed through a subsequent generation task. We surmise that this procedure might be especially useful in areas in which short procedures are needed, such as when special populations are tested (e.g., patients, children or elderly people) or when the neural bases of implicit learning are being investigated through neurophysiological measures.     

Sleep deprivation increases threat beliefs in human fear conditioning

Sleep disturbances and anxiety disorders exhibit high comorbidity levels, but it remains unclear whether sleep problems are causes or consequences of increased anxiety. To experimentally probe the aetiological role of sleep disturbances in anxiety, we investigated in healthy participants how total sleep deprivation influences fear expression in a conditioning paradigm. In a fear conditioning procedure, one face stimulus (conditioned stimulus [CS+]) was paired with electric shock, whereas another face stimulus was not (unpaired stimulus [CS?]). Fear expression was tested the next morning using the two face stimuli from the training phase and a generalization stimulus (i.e. a morph between the CS+ and CS? stimuli). Between fear conditioning and test, participants were either kept awake in the laboratory for 12 hr (n = 20) or had a night of sleep at home (n = 20). Irrespective of stimulus type, subjective threat expectancies, but not skin conductance responses, were enhanced after sleep deprivation, relative to regular sleep. These results suggest that sleep disturbances may play a role in anxiety disorders by increasing perceived threat.     

Sensory information in perceptual-motor sequence learning: visual and/or tactile stimuli

Sequence learning in serial reaction time (SRT) tasks has been investigated mostly with unimodal stimulus presentation. This approach disregards the possibility that sequence acquisition may be guided by multiple sources of sensory information simultaneously. In the current study we trained participants in a SRT task with visual only, tactile only, or bimodal (visual and tactile) stimulus presentation. Sequence performance for the bimodal and visual only training groups was similar, while both performed better than the tactile only training group. In a subsequent transfer phase, participants from all three training groups were tested in conditions with visual, tactile, and bimodal stimulus presentation. Sequence performance between the visual only and bimodal training groups again was highly similar across these identical stimulus conditions, indicating that the addition of tactile stimuli did not benefit the bimodal training group. Additionally, comparing across identical stimulus conditions in the transfer phase showed that the lesser sequence performance from the tactile only group during training probably did not reflect a difference in sequence learning but rather just a difference in expression of the sequence knowledge.

Image analysis and artificial intelligence in infectious disease diagnostics

BackgroundMicrobiologists are valued for their time-honed skills in image analysis, including identification of pathogens and inflammatory context in Gram stains, ova and parasite preparations, blood smears and histopathologic slides. They also must classify colony growth on a variety of agar plates for triage and assessment. Recent advances in image analysis, in particular application of artificial intelligence (AI), have the potential to automate these processes and support more timely and accurate diagnoses.ObjectivesTo review current AI-based image analysis as applied to clinical microbiology; and to discuss future trends in the field.SourcesMaterial sourced for this review included peer-reviewed literature annotated in the PubMed or Google Scholar databases and preprint articles from bioRxiv. Articles describing use of AI for analysis of images used in infectious disease diagnostics were reviewed.ContentWe describe application of machine learning towards analysis of different types of microbiologic image data. Specifically, we outline progress in smear and plate interpretation as well as the potential for AI diagnostic applications in the clinical microbiology laboratory.ImplicationsCombined with automation, we predict that AI algorithms will be used in the future to prescreen and preclassify image data, thereby increasing productivity and enabling more accurate diagnoses through collaboration between the AI and the microbiologist. Once developed, image-based AI analysis is inexpensive and amenable to local and remote diagnostic use.     

Pavlovian aversive and appetitive odor conditioning in humans: subjective, peripheral, and electrocortical changes

The effects of presynaptic guanosine-5'-O-(3-thio)triphosphate (GTPγS) on GABAergic inhibitory postsynaptic currents (IPSCs) were studied in cultured hippocampal neurons using whole-cell recordings. Inclusion of GTPγS (0.5–1 mM) in the presynaptic electrode reduced both the amplitude and paired-pulse depression of IPSCs, indicating that the probability of GABA-release had been reduced. Presynaptic GTPγS increased the depression of IPSCs by the GABA_B-receptor-agonist baclofen (10 μM), and the effect of baclofen was poorly reversible after washing. Stimulation of the GABAergic neuron at 80 Hz for 1 s was accompanied by tetanic depression of the IPSCs by 52±6% and was followed by post-tetanic potentiation (PTP), reaching a peak value of 71±21% and lasting about 100 s. IPSCs evoked after tetanic stimulation were depressed and PTP was absent when tetanic stimulation was applied within 3 min after starting injection of GTPγS into the presynaptic neuron. At longer times, basal release underlying a single IPSC was depressed. This affected the ratios recorded in response to tetanic stimulations such that tetanic depression was abolished, while PTP increased to 117±34%. In conclusion, GTPγS reduces the probability of GABA-release in both a use- and time-dependent manner, most likely through an inhibitory action on presynaptic Ca²⁺-influx through voltage-gated Ca²⁺ channels or an interaction with small GTP-binding proteins in the nerve terminals. Electronic Publication