基于深度学习的心脏核磁共振图像自动分割算法

心脏核磁共振成像技术由于其无电离辐射的优点已成为医疗诊断中的主要手段。对左心室、右心室以及左心肌进行准确的分割与识别是心脏手术前的重要一步,手动分割心脏结构耗时且易出错,因此自动分割双心室与心肌至关重要。提出了一种能充分利用心脏图像信息的多尺度特征融合U型神经网络MFF U-Net。首先,选择以U-Net++作为网络基本框架。其次,为了提高特征复用率,解决网络深度增加导致的过拟合问题,在U-Net++的编码部分提出了密集残差模块,使得网络在下采样过程中学习到更多有用特征。此外,在解码部分,为了使网络的分割结果更加符合目标器官之间的物理特征,用多个卷积核来扩大感受野并利用长距离依赖模块共享全局上下文信息,使得网络在编码还原的过程中尽可能地获取到目标器官之间的关系信息,从而使得分割结果更为精准。最后,考虑到双心室与左心肌的连贯性与唯一性,还添加了获取最大连通域与填充细小孔洞的后处理操作。采用的实验数据为ACDC心脏分割挑战数据集,其包含150位志愿者收缩期末期与舒张期末期的短轴心脏磁共振图像。在该数据集的测试集上进行验证,并通过在线提交的方式获取实验结果。实验结果表明,相较于其他算法,所提出的算法能够有效地分割目标器官,特别是舒张期末期的Dice系数分别达到了左心室0.96、右心室0.94和左心肌0.89,收缩期末期的分割精度达到了0.87,0.86和0.89。     

深度学习的快速磁共振成像及欠采样轨迹设计

目的 快速成像一直是磁共振成像（MRI）技术中的焦点之一,现有多通道并行成像和部分k空间数据重建都是通过减少梯度编码步数来降低数据的获取时间,两者结合起来更能有效地提高扫描速度。然而,在欠采样倍数加高的情况下,依然有很严重的混叠伪影,因此研究一种在保证成像精度的前提下加快成像速度的方法尤为重要。方法 基于卷积神经网络的磁共振成像（CNN-MRI）方法利用大量现有的全采样多通道数据的先验信息,设计并线下训练一个深度卷积神经网络,学习待重建图像与全采样图像之间的映射关系,从而在线上成像时,欠采样所丢失数据能被训练好的网络进行预测。本文探讨了对于深度学习磁共振成像的可选择性欠采样方式,提出了一种新的欠采样轨迹方案。为了判断本文方法的性能,用峰值信噪比（PSNR）、结构相似度（SSIM）以及均方根差（RMSE）来作为衡量的指标。结果 实验结果表明,所提出欠采样方案的综合性能要优于传统欠采样轨迹,PSNR要高出12 dB,SSIM高出近0.1,RMSE要降低0.020.04左右。此外重建结果还与经典的并行重建方法GRAPPA（geneRalized autocalibrating partially parallel acquisitions）、SPIRiT（iterative self-consistent parallel imaging reconstruction from arbitrary k-space）以及SAKE（simultaneous autocalibrating and k-space estimation）作比较,从视觉效果以及各项量化指标得出本文方法能重建出更准确的结果,并且重建速度要快5倍以上。结论 深度学习方法能很好地在线下训练时从大量数据集中提取并学习到有价值的先验信息,所以在线上测试时能在较短时间内重建出优于经典算法的高质量结果;提出的1维低频汉明滤波欠采样方案则有利于提升该网络的性能。     

基于深度学习与结构磁共振成像的抑郁症辅助诊断

抑郁症是致残率和发病率最高的疾病之一,全球约有3亿人正遭受着抑郁症的困扰.然而,目前并没有有效的生物特征和临床方法能够帮助医生对抑郁症进行准确的诊断.针对此任务,本文将计算机视觉领域的前沿深度学习模型进行优化与适配,应用于抑郁症的辅助诊断,并在此基础上引入迁移学习,取得了很好的效果.实验结果表明,同前沿算法模型相比,本...     

基于深度学习的心脏磁共振影像超分辨率前沿进展

心脏为人体血液流动提供动力,是人体血液循环系统的重要组成部分。受人口老龄化影响,心脏病诊疗已成为重大公共健康话题。非侵入式活体心脏成像对心脏疾病的检测、诊断与治疗意义重大。然而,受活体心跳影响,成像扫描时间与心脏影像分辨率成为难以调和的矛盾。为缓和这一矛盾,基于快速扫描获得的低分辨率影像重建出心脏高分辨率影像的超分辨率(super-resolution, SR)重建技术成为研究热点。深度学习技术在医学影像处理领域中展现出强大生命力,基于深度学习的SR技术因其强大的学习能力与数据驱动性,在心脏影像SR重建领域中表现出明显优于传统方法的性能。目前领域内前沿成果较多,但缺少对领域现状进行总结、对未来发展进行展望的综述性文献。因此,本文对领域内现状进行梳理总结,挑选出代表性方法,分析方法特性,总结文献中心脏影像数据来源与规模,给出常用的评价指标,以及模型得出的性能评价结论。分析发现,基于深度学习的心脏SR重建技术取得了较大进展,但在运动伪影抑制、模型简化程度与时间性能方面仍有进步空间。此外,现有模型基本完全依靠网络强大的表达能力,鲜有临床先验知识的引入。最后,模型间性能对比相对较少,且领域内缺...     

A combined deep learning method for internet car evaluation

Neural Computing and Applications - In recent years, the Internet has become a trend in the development of the global automotive industry. Numerous Internet companies have joined the automobile...     

Player target tracking and detection in football game video using edge computing and deep learning

The purpose is to explore the player detection and motion tracking in football game video based on edge computing and deep learning (DL), thus improving the detection effect of player trajectory in different scenes. First, the basic technology of player target tracking and detection task is analyzed based on the Histograms of Oriented Gradients feature. Then, the neural network structure in DL is combined with the target tracking method to improve the miss detection problem of the Faster R-CNN (FRCN) algorithm in detecting small targets. Edge computing places massive computing nodes close to the terminal devices to meet the high computing and low latency requirements of DL on edge devices. After the occlusion problem in the football game is analyzed, the optimized algorithm is applied to the public dataset OTB2013 and the football game dataset containing 80 motion trajectories. After testing, the target tracking accuracy of the improved FRCN is 89.1%, the target tracking success rate is 64.5%, and the running frame rate is still about 25 fps. The high confidence of FRCN algorithm also avoids template pollution. In the ordinary scene, the FRCN algorithm basically does not lose the target. The area under curve value of the proposed FRCN algorithm decreases slightly in the scene where the target is occluded. The FRCN algorithm based on DL technology can achieve the target tracking of players in football game video and has strong robustness to the situation of players occlusion. The designed target detection algorithm is applied to the football game video, which can better analyze the technical characteristics of players, promote the development of football technology, bring different viewing experiences to the audience, drive the development of economic products derived from football games, and promote the dissemination and promotion of football.

     

A deep learning approach for the forensic evaluation of sexual assault

Despite the existence of patterns able to discriminate between consensual and non-consensual intercourse, the relevance of genital lesions in the corroboration of a legal rape complaint is currently under debate in many countries. The testimony of the physicians when assessing these lesions has been questioned in court due to several factors (e.g., a lack of comprehensive knowledge of lesions, wide spectrum of background area, among others). Therefore, it is relevant to provide automated tools to support the decision process in an objective manner. In this work, we evaluate the performance of state-of-the-art deep learning architectures for the forensic assessment of sexual assault. We propose a deep architecture and learning strategy to tackle the class imbalance on deep learning using ranking. The proposed methodologies achieved the best results when compared with handcrafted feature engineering and with other deep architectures.     

融合深度置信网络与与核极限学习机算法的核磁共振测井储层渗透率预测方法

由于低孔低渗储层孔隙结构较为复杂,现有核磁共振（NMR）测井渗透率模型对于低孔低渗储层预测精度不高。为此,提出一种融合深度置信网络（DBN）算法与核极限学习机（KELM）算法的渗透率预测方法。该方法首先对DBN模型进行预训练,然后将KELM模型作为预测器放置在训练好DBN模型后,利用训练数据进行有监督的训练,最终形成深度置信-核极限学习机（DBKELMN）模型。考虑到该模型需充分利用反映孔隙结构的横向弛豫时间谱信息,将离散化后的核磁共振测井横向弛豫时间谱作为输入,渗透率作为输出,确定NMR测井横向弛豫时间谱与渗透率的函数关系,并基于该函数关系对储层渗透率进行预测。实例应用表明,融合DBN算法与KELM算法的渗透率预测方法是有效的,预测样本的平均绝对误差（MAE）较斯伦贝谢道尔研究中心（SDR）模型降低了0.34。融合DBN算法与KELM算法的渗透率预测方法可提高低孔渗储层渗透率预测精度,可应用于油气田勘探开发。     

深度学习下主流染色体分类算法的性能评估

目的 染色体分类是医学影像处理的具体任务之一,最终结果可为医生提供重要的临床诊断信息,在产前诊断中起着重要作用。深度学习由于强大的特征表达能力在医学影像领域得到了广泛应用,但是基于深度学习的大部分染色体分类算法都是在轻量化私有数据库上得到的不同水准的分类结果,难以客观评估不同算法间的优劣,导致缺乏对算法的临床筛选标准,因此迫切需要在大规模数据库上对不同算法开展基于同样数据级的性能评估,以获取具有客观可对比性的性能数据,这对于科研成果的转化具有重要意义。方法 本文基于广东省妇幼保健院提供的染色体数据,构建了包含126 453条染色体的临床数据库,精选6个主流染色体分类模型在该数据库上展开对比实验与性能评估。结果 在本文构建的大规模染色体临床数据库上,实验和分析发现,参评模型分类准确率均达到92%以上,其中MixNet模型提升后分类效果最好,为98.92%。即使分类性能落后的模型在本数据集上训练也得到明显改善,准确率从86.7%提升至92.09%,相比早期报告的性能提升了5.39%。结论 开展实证研究实验发现,数据库规模大小是影响染色体分类算法能否取得理想分类效果的重要因素之一。对于染色体...     

Performance evaluation of edge-computing platforms for the prediction of low temperatures in agriculture using deep learning

The Journal of Supercomputing - The Internet of Things (IoT) is driving the digital revolution. AlSome palliative measures aremost all economic sectors are becoming “Smart” thanks to...     

Degradation evaluation of lateral story stiffness using HLA-based deep learning networks

Hysteresis loop analysis (HLA) has proven an effective indicator of damage detection in civil engineering structural health monitoring (SHM). In this paper, the histogram of stiffness (HOS) features are extracted from segregated half cycles of hysteresis loops reconstructed from measured response. A deep learning network (DLN) is proposed with the use of the HOS to classify the damage index (DI) based on stiffness degradation for damage identification. Training data are obtained using numerical simulations of 30,000 realistic, randomly created hysteresis loops, including a wide range of typical linear and nonlinear structural behaviours. Performance of the trained DLN model is assessed using both 1800 additional simulated 3-story “virtual” buildings and experimental data from a 3-story full-scale real building. Results are compared to the validated HLA method.Validation on simulated virtual building data yields prediction accuracy for 97.2% and 91.6% samples without and with 10% added noise, respectively. The comparison shows a good match of trend and percentage stiffness drop between DLN and HLA identification with the average difference for all cases within 1.1–4.6%, indicating a good accuracy of the proposed DLN prediction model for real structures. The overall results show its potential to provide a rapid, and real-time alarm or other notice on damage states and mitigation to emergency response using DLN and thus without detailed engineering analysis.     

Ultrasonic evaluation of fetal lung development using deep learning with graph

BackgroundThe neonatal respiratory morbidity that was primarily caused by the immaturity of the fetal lung is an important clinical issue in close relation to the morbidity and mortality of the fetus. In clinics, the amniocentesis has been used to evaluate the fetal lung maturity, which is time-consuming, costly and invasive. As a non-invasive means, ultrasonography has been explored to quantitatively examine the fetal lung in the past decades. However, existing studies required the contour of the fetal lung which was delineated manually. This may lead to significant inter- and intra-observer variations.MethodsWe proposed a deep learning model for automated fetal lung segmentation and measurement, which was constructed combined U-Net with Graph model and pre-trained Vgg-16 network. The graph connection would extract stable feature for final segmentation and pre-trained method could speed up convergence.The model was trained with 3500 datasets augmented from 250 ultrasound images with both the fetal lung and heart delineated manually, and tested on 50 ultrasound images. In addition, the correlation between the size of fetal lung/heart as delineated by the model with gestational age was analyzed.ResultsThe fetal lung and cardiac area were segmented automatically with the accuracy, average Intersection over Union(IoU), sensitivity and precision being 0.991, 0.818, 0.909 and 0.888, respectively. In addition, the size of fetal lung/heart was well correlated with the gestational age, demonstrating good potentials for assessing the fetal development.ConclusionsThis study proposed a new robust method for automatic fetal lung segmentation in ultrasound images using Vgg16-GCN-UNet. Our proposed method could be utilized potentially not only to improve existing research in quantitative analyzing the fetal lung using ultrasound imaging technology, but also to alleviate the labor of the clinicians in routine measurement of the fetal lung/cardiac.     

深度学习图像数据增广方法研究综述

数据作为深度学习的驱动力，对于模型的训练至关重要。充足的训练数据不仅可以缓解模型在训练时的过拟合问题，而且可以进一步扩大参数搜索空间，帮助模型进一步朝着全局最优解优化。然而，在许多领域或任务中，获取到充足训练样本的难度和代价非常高。因此，数据增广成为一种常用的增加训练样本的手段。本文对目前深度学习中的图像数据增广方法进行研究综述，梳理了目前深度学习领域为缓解模型过拟合问题而提出的各类数据增广方法，按照方法本质原理的不同，将其分为单数据变形、多数据混合、学习数据分布和学习增广策略等4类方法，并以图像数据为主要研究对象，对各类算法进一步按照核心思想进行细分，并对方法的原理、适用场景和优缺点进行比较和分析，帮助研究者根据数据的特点选用合适的数据增广方法，为后续国内外研究者应用和发展研究数据增广方法提供基础。针对图像的数据增广方法，单数据变形方法主要可以分为几何变换、色域变换、清晰度变换、噪声注入和局部擦除等5种；多数据混合可按照图像维度的混合和特征空间下的混合进行划分；学习数据分布的方法主要基于生成对抗网络和图像风格迁移的应用进行划分；学习增广策略的典型方法则可以按照基于元学习和基于强化学习进行分类。目前，数据增广已然成为推进深度学习在各领域应用的一项重要技术，可以很有效地缓解训练数据不足带来的深度学习模型过拟合的问题，进一步提高模型的精度。在实际应用中可根据数据和任务的特点选择和组合最合适的方法，形成一套有效的数据增广方案，进而为深度学习方法的应用提供更强的动力。在未来，根据数据和任务基于强化学习探索最优的组合策略，基于元学习自适应地学习最优数据变形和混合方式，基于生成对抗网络进一步拟合真实数据分布以采样高质量的未知数据，基于风格迁移探索多模态数据互相转换的应用，这些研究方向十分值得探索并且具有广阔的发展前景。     

Application of embedded wearable devices in football training injury prevention

Knee trauma is a significant physical concern for footballers and members of various sports around the world. There is no extended test from the lab to demonstrate the potential benefits of biomechanics and developers engaging in intensive training to prevent knee injuries. Because of high-speed and full contact football peaks matchups and mixed practice injuries. Although injuries can be misdiagnosed, such electric shock-inducing injuries are usually normal. Electric football is suitable for anyone who brings enemies against the ground or on the ground, no matter how their bodies tend to be injured anywhere in the focus of defensive hardware pay. For these developers, there are not many global choices that have already been evaluated as to whether the online environment has added infectious behavior and biomechanical methods in the real environment. After comparing the rate and severity of the injury mediated benchmark group. Changes, if any, are the social area. Cycle Rating: mentor, transport factors, and potential manageability.     

An optimal deep learning framework for multi-type hemorrhagic lesions detection and quantification in head CT images for traumatic brain injury

Applied Intelligence - Traumatic Brain Injury (TBI) could lead to intracranial hemorrhage (ICH), which has now been identified as a major cause of death after trauma if it is not adequately...     

Advances in deep learning for multimodal fusion and alignment

Multimedia Tools and Applications -     

染色体核型分析深度学习方法综述

染色体核型分析是细胞遗传学领域重要的实验技术,并逐步在包括生殖医学在内的诸多现代临床领域和科学研究方面得到广泛应用,但即使是经验丰富的细胞遗传学家也需要大量时间才能完成染色体核型分析。基于传统方法的染色体核型自动化分析方法精度较低,仍需要细胞遗传学家花费大量时间、精力校正。目前基于深度学习的染色体核型自动分析方法成果较多,但缺乏对该领域现状的总结、对未来发展的展望等。因此,本文对基于深度学习的染色体核型自动分析方法进行综述,归纳总结了现有的研究分析任务,挑选了具有代表性的方法并梳理解决方案,展望了未来发展方向。通过整理发现,基于深度学习的染色体核型自动化分析方法取得了很多成果,但仍存在一些问题。首先,已有的中文综述性工作仅集中于某一子领域或者调研不够全面和深入。其次,染色体核型分析任务与临床紧密结合,受各种因素制约,任务类型繁多,解决方案复杂,难以窥斑见豹。最后,现有方法主要集中于染色体分类和染色体分割任务,而诸如染色体计数、染色体预处理等任务研究成果较少,需要厘清问题,吸引更多研究人员关注。综上所述,基于深度学习的染色体核型自动分析方法仍有较大发展空间。     

有监督深度学习的优化方法研究综述

随着大数据的普及和算力的提升,深度学习已成为一个热门研究领域,但其强大的性能过分依赖网络结构和参数设置。因此,如何在提高模型性能的同时降低模型的复杂度,关键在于模型优化。为了更加精简地描述优化问题,本文以有监督深度学习作为切入点,对其提升拟合能力和泛化能力的优化方法进行归纳分析。给出优化的基本公式并阐述其核心;其次,从拟合能力的角度将优化问题分解为3个优化方向,即收敛性、收敛速度和全局质量问题,并总结分析这3个优化方向中的具体方法与研究成果;从提升模型泛化能力的角度出发,分为数据预处理和模型参数限制两类对正则化方法的研究现状进行梳理;结合上述理论基础,以生成对抗网络（generative adversarial network,GAN）变体模型的发展历程为主线,回顾各种优化方法在该领域的应用,并基于实验结果对优化效果进行比较和分析,进一步给出几种在GAN领域效果较好的优化策略。现阶段,各种优化方法已普遍应用于深度学习模型,能够较好地提升模型的拟合能力,同时通过正则化缓解模型过拟合问题来提高模型的鲁棒性。尽管深度学习的优化领域已得到广泛研究,但仍缺少成熟的系统性理论来指导优化方法的使用,...     

An unsupervised learning method with a clustering approach for tumor identification and tissue segmentation in magnetic resonance brain images

Malignant and benign types of tumor infiltrated in human brain are diagnosed with the help of an MRI scanner. With the slice images obtained using an MRI scanner, certain image processing techniques are utilized to have a clear anatomy of brain tissues. One such image processing technique is hybrid self-organizing map (SOM) with fuzzy K means (FKM) algorithm, which offers successful identification of tumor and good segmentation of tissue regions present inside the tissues of brain. The proposed algorithm is efficient in terms of Jaccard Index, Dice Overlap Index (DOI), sensitivity, specificity, peak signal to noise ratio (PSNR), mean square error (MSE), computational time and memory requirement. The algorithm proposed through this paper has better data handling capacities and it also performs efficient processing upon the input magnetic resonance (MR) brain images. Automatic detection of tumor region in MR (magnetic resonance) brain images has a high impact in helping the radio surgeons assess the size of the tumor present inside the tissues of brain and it also supports in identifying the exact topographical location of tumor region. The proposed hybrid SOM-FKM algorithm assists the radio surgeon by providing an automated tissue segmentation and tumor identification, thus enhancing radio therapeutic procedures. The efficiency of the proposed technique is verified using the clinical images obtained from four patients, along with the images taken from Harvard Brain Repository.     

Applying learning behavioral Petri nets to the analysis of learning behavior in web-based learning environments

This paper proposes the learning behavioral Petri nets (LBPN) to model learning behavior in web-based environments. Fully useful records of learning behaviors must contain their expended time and corresponding contents. Therefore, the LBPN extends the colored tokens of colored Petri nets to identify learners and learning contents, and raises the time variable to represent diverse learning times for individual learners. To verify the viability of the LBPN, this paper also proposes a LBPN-based learning behavioral model to simulate a situation in which many learners participate in an e-learning course, and then to generate their behavioral patterns. The experimental results illustrated in this paper confirm that (1) the generated behavioral pattern based on the LBPN-based model is very close to actual data, (2) the time and cost spent to verify the effectiveness of an ITS is substantially reduced, (3) adequate testing data for estimating the performance and accuracy of an ITS is easily acquired, and (4) the LBPN-based model can be built to recommend appropriate learning contents and to accomplish adaptive learning.