基于遗传算法反求耳结构弹性模量

本研究通过力学反问题原理,利用已知的位移求解耳结构弹性模量。随机产生遗传算法的初始种群,使用自编的Matlab算法程序,对初始种群进行遗传迭代计算,把已知的目标位移与种群位移的均方差作为目标函数,以目标函数值最小控制迭代进化的方向。通过耳结构的鼓膜凸和镫骨底板2个控制位移以及砧镫关节周围的8个控制位移这两个算例求解正常砧镫关节的弹性模量,并使用耳结构的鼓膜凸和镫骨底板2个控制位移求解病变砧镫关节的弹性模量。结果表明,使用基于遗传算法的反问题方法计算耳结构的弹性模量是可行的,并且具有稳定性和不受结构力学性能影响的特点,相对误差分别为0.05%和0.2%、0.03%,可为临床病变耳提供有效的力学参数。     

人耳鼓膜病变数值分析

目的研究鼓膜厚度和硬度对人耳传声的影响。方法利用CT获取志愿者耳部结构临床资料,使用Matlab软件提取相关结构的边界,将边界文件导入ANSYS建立人耳结构数值有限元模型。结果利用本文人耳数值模型,在外耳道口施加105dB声压,进行200～8000Hz频率范围的谐响应分析。以此研究在鼓膜病变情况下,鼓膜和镫骨底板位移幅值的变化规律。结论用数值方法解释了鼓膜病变对传声的影响,为鼓膜修补提供了力学参考。     

Stress analysis between implant prostheses with different moduli and surrounding bones北大核心

背景:应力遮挡效应会导致植入假体修复骨缺损手术失败,其主要原因是由于植入假体的弹性模量大于骨组织弹性模量。目的:分析植入假体弹性模量对应力分布的影响,寻求消除应力遮挡现象的方法。方法:通过CT扫描的方式获取实验犬与人体骨组织的模型,分别对其优化后进行梯度赋值,建立较为可靠的骨骼力学模型,并与植入假体组合后进行有限元仿真。首先,通过对比格犬骨骼模型和人体骨骼模型及其对应的植入假体进行有限元仿真,模拟了不同弹性模量对植入假体修复术后的应力和位移分布情况;其次,分析了较小弹性模量差仍会形成应力遮挡现象的原因,建立了骨骼模型及植入假体模型,确立了材料属性赋予方法;最后,验证了该模型及材料属性赋予方法的可行性,并通过随机选取受力点的方式,定量分析植入假体弹性模量与骨骼弹性模量之间的关系对应力遮挡形成的影响。结果与结论:通过梯度赋值法建立与骨骼力学性质更加接近的实验犬骨骼模型和人体骨骼模型,该方法重建的力学模型与真实骨骼的力学性质更为接近;通过有限元仿真力学测试证明,不同弹性模量植入假体对假体本身与周围骨骼间相对位移的影响较小;另外量化弹性模量对假体植入骨骼后对应力分布的影响,可为后续的相关研究提供帮助。     

基于B P神经网络的脑肿瘤MRI图像分割

利用BP神经网络技术对MR脑肿瘤图像中的肿瘤区域与正常组织区域进行分割,以辅助医疗诊断与治疗。首先,人工分割出部分影像中的肿瘤组织与正常组织作为已知样本;其次,在BP神经网络模型中输入已知样本中进行训练;最后,用训练好的BP神经网络处理其他脑肿瘤图像。BP神经网络能够有效分割MR脑肿瘤图像,辨别出肿瘤与周围正常组织的差异,但模糊区域也常被误判为肿瘤。因此,本研究提出进一步对模糊区域样本进行针对性训练与特殊的滤波处理,所得结果有较大改进。BP神经网络能有效地进行脑肿瘤MRI图像分割,但在使用时仍需正确选择输入样本的区域和范围并结合特殊的滤波处理。     

基于粒子群和反向传播神经网络的近红外光无创血糖检测方法研究EI北大核心CSCD

现有的近红外无创血糖检测模型研究大多数关注的是近红外吸光度与血糖浓度之间的关系,但没有考虑人体生理状态对血糖浓度的影响。为了提升血糖预测模型性能,本文采用了粒子群优化算法(PSO)对反向传播(BP)神经网络的结构参数进行训练,并引入了收缩压、脉率、体温以及1550 nm吸光度作为血糖浓度预测模型的输入变量,采用BP神经网络作为预测模型。为解决传统BP神经网络容易陷入局部最优的问题,本文提出了一种基于PSO-BP的混合模型。结果表明,训练得到的PSO-BP模型预测效果优于传统的BP神经网络。十折交叉验证预测均方根误差和相关系数分别为0.95 mmol/L和0.74;克拉克误差网格分析结果表明,模型预测结果落入A区域的比例为84.39%,落入B区域的比例为15.61%,均满足临床要求。该模型可以快速地测量血糖浓度,且具相对较高的精度。     

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BP神经网络建立的川崎病早期诊断模型

为早期诊断川崎痛,应用BP神经网络原理建立川崎病的诊断模型.以156例川崎病与非川崎病患者的体温、皮疹、口腔黏膜改变、实验室检查结果等9项指标等作为BP神经网络的输入参数,在MATLAB7程序中对其中随机抽取的90例学习样本进行训练并建模.以剩余的66例作为测试样本进行预测,结果表明该模型对川崎病和非川崎病的预测准确率...     

基于GA-BP神经网络进行呼吸运动预测的研究

放射治疗是胸腹部肿瘤的常用治疗手段,但由于病人在治疗过程中的呼吸使胸腹部肿瘤产生运动位移降低了放射治疗的质量。人们希望在治疗的过程中对目标进行运动估计,以减少呼吸运动在治疗过程中的影响。BP人工神经网络由于其良好的非线性逼近特性被用于呼吸运动估计,可以较好地反映呼吸运动的发展趋势;但BP神经网络本质是梯度下降法,容易陷入局部最优。利用遗传算法(GA)和BP网络相结合的方法,遗传算法具有良好的全局搜索能力,弥补BP神经网络易于陷入局部最优的缺点。通过对9例呼吸运动信号进行对比实验,结果表明用GA-BP神经网络的预测精度高于单纯使用BP神经网络的预测精度。     

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。     

基于神经网络的环孢素血药浓度预测

通过研究肾移植病人环孢素A血药浓度的不同影响因子,分别采用广义神经网络、Elman网络、BP神经网络等模型,对环孢素A血药浓度进行预测.结果显示三种网络模型都达到了较好的预测结果,其中BP网络的预测结果最好,平均相对误差为19.94%,Elman网络的平均相对误差为21.39%,广义神经网络的平均相对误差为25.93%.说明将神经网络应用于预测CsA血药浓度是可行的,其结果可以作为临床CsA的个体化给药的参考.     

Elastic modulus and stress relationships in stretched and shortened frog sartorii.

The longitudinal dynamic elastic modulus (delta stress/delta strain) of 11 sartorius muscles (Rana pipiens; 3 degrees C) was measured at static strains, 0.57 less than L/Lo less than 1.53. Pseudorandom white-noise displacements less than +/-0.031% Lo (mean, peak to peak) were imposed on tetanically stimulated and resting muscles to obtain the isometric moduli. The active elastic modulus at each length was determined as the difference between the low-frequency (15-80 Hz) asymptotes of the tetanic and resting modulus functions. Resting moduli were found to increase with stretch at a declining rate, suggesting that some resting elasticity is attributable to active crossbridges. For isometric, tetanically stimulated muscles above L/Lo = 1.0, the ratio dynamic elastic modulus/active stress was nearly constant (65.4); the data predict zero modulus at a stretched length corresponding to zero active stress. Hence, the modulus per crossbridge exhibits sarcomere length invariance. On the other hand, active muscles below L/Lo = 0.77 manifest a significant (P less than or equal to 0.05) additional modulus beyond that found at the same active stress in the stretched muscle. Sarcomeric structural rearrangements are suggested as a possible source of this additional modulus.     

弹性模量与表观密度的分段函数关系用于股骨近端的结构模拟

目前，在用功能适应性理论与有限元相结合，进行计算机数值模拟骨再造时，大多数研究者对于不同的表观密度，都是将表观密度和弹性模量的关系取为一种形式。由于松质骨的各向异性，其结构形式的多样性、复杂性，认为用一个统一的模型来描述其结构性能存在一定的问题。本研究基于松质骨胞元模型理论，提出与实际更加接近的松质骨弹性模量与表观密度的分段函数关系，使得弹性模量与表观密度关系的确立有一定的理论基础。利用Mullender等提出的骨自我组织控制过程微分方程来控制骨再造过程，与有限元相结合，将这种松质骨的弹性模量与表观密度的分段函数关系应用到股骨近端结构的模拟预测中，不但能根据模拟得到的内部密度分布给出其内部结构的具体胞元结构形式，而且可以大大提高迭代的收敛速度，因此是一项具有一定的理论意义和实际应用价值的工作。     

Experimental and numerical analyses of local mechanical properties measured by atomic force microscopy for sheared endothelial cells

Local mechanical properties were measured for bovine endothelial cells exposed to shear stress using an atomic force microscopy (AFM), and the AFM indentations were simulated using a finite element method (FEM) to determine the elastic modulus. After exposure to shear stress, the endothelial cells showed marked elongation and orientation in the flow direction, together with significant decrease in the peak cell height. The applied force-indentation depth curve was obtained at seven different locations on the major axis of the cell surface and quantitatively expressed by the quadratic equation. The elastic modulus was determined by comparison of the experimental and numerical results. The modulus using our FEM model significantly became higher from 12.2+/-4.2 to 18.7+/-5.7 kPa with exposure to shear stress. Fluorescent images showed that stress fibers of F-actin bundles were mainly formed in the central portion of the sheared cells. The significant increase in the modulus may be due to this remodeling of cytoskeletal structure. The moduli using the Hertz model are 0.87+/-0.23 and 1.75+/-0.43 kPa for control and sheared endothelial cells respectively. This difference can be attributable to the differences in approximation functions to determine the elastic modulus. The elastic modulus would contribute a better understanding of local mechanical properties of the cells.     

In vivo areal modulus of elasticity estimation of the human tympanic membrane system: modelling of middle ear mechanical function in normal young and aged ears

The quasi-static elastic properties of the tympanic membrane system can be described by the areal modulus of elasticity determined by a middle ear model. The response of the tympanic membrane to quasi-static pressure changes is determined by its elastic properties. Several clinical problems are related to these, but studies are few and mostly not comparable. The elastic properties of membranes can be described by the areal modulus, and these may also be susceptible to age-related changes reflected by changes in the areal modulus. The areal modulus is determined by the relationship between membrane tension and change of the surface area relative to the undeformed surface area. A middle ear model determined the tension-strain relationship in vivo based on data from experimental pressure-volume deformations of the human tympanic membrane system. The areal modulus was determined in both a younger (n = 10) and an older (n = 10) group of normal subjects. The areal modulus for lateral and medial displacement of the tympanic membrane system was smaller in the older group (mean = 0.686 and 0.828 kN m(-1), respectively) compared to the younger group (mean = 1.066 and 1.206 kN m(-1), respectively), though not significantly (2p = 0.10 and 0.11, respectively). Based on the model the areal modulus was established describing the summated elastic properties of the tympanic membrane system. Future model improvements include exact determination of the tympanic membrane area accounting for its shape via 3D finite element analyses. In vivo estimates of Young's modulus in this study were a factor 2-3 smaller than previously found in vitro. No significant age-related differences were found in the elastic properties as expressed by the areal modulus.     

Characterization of cellular elastic modulus using structure based double layer model

The mechanical characterization of cells is important for understanding cellular behavior and physiological functions. We used atomic force microscopy (AFM) to obtain a force-displacement curve and estimate the elastic modulus of hepatocellular carcinoma cells (HEP-G2) utilizing both linear Hertz-Sneddon (HS) and non-linear elastic models. In order to overcome the limitations of HS model, which assumes a linear homogeneous cell body, a cell is modeled as a double-layered body with an outer cytoplasmic layer made mostly of interconnected fibers of cytoskeleton proteins and a nucleus. By disrupting all cytoskeletal protein networks, we estimate the elastic modulus of the core nucleus using FEM for a single ellipsoid. Based on the nucleic modulus and cellular dimensions found by 3D confocal imaging, we develop a novel double-layered cellular (DLC) finite element model. The DLC model provides a more reliable estimate of the elastic modulus of the cell than conventionally used HS model and correlates closely with experimental results.     

RBF networks for source localization in quantitative electrophysiology

The backpropagation neural network methods have been proposed recently to solve the inverse problem in quantitative electrophysiology. A major advantage of the technique is that once a neural network is trained, it no longer requires iterations or access to sophisticated computations. We propose to use RBF networks for source localization in the brain, and systematically compare their performance to those of Levenberg-Marquardt (LM) algorithms. We show the use of two types of Radial Basis Function Networks (RBF) network: a classic network with fixed number of hidden layer neurons and an improved network, Minimal Resource Allocation Network (MRAN), recently proposed by one of the authors, capable for dynamically configuring its structure so as to obtain a compact topology to match the data presented to it.     

Quantitative elasticity imaging: what can and cannot be inferred from strain images

We examine the inverse problem associated with quantitative elastic modulus imaging: given the equilibrium strain field in a 2D incompressible elastic material, determine the elastic stiffness (shear modulus). We show analytically that a direct formulation of the inverse problem has no unique solution unless stiffness information is known a priori on a sufficient portion of the boundary. This implies that relative stiffness images constructed on the assumption of constant boundary stiffness are in error, unless the stiffness is truly constant on the boundary. We show further that using displacement boundary conditions in the forward incompressible elasticity problem leads to a nonunique inverse problem. Indeed, we give examples in which exactly the same strain field results from different elastic modulus distributions under displacement boundary conditions. We also show that knowing the stress on the boundary can, in certain configurations, lead to a well-posed inverse problem for the elastic stiffness. These results indicate what data must be taken if the elastic modulus is to be reconstructed reliably and quantitatively from a strain image.     

超声弹性成像中的逆问题求解方法

超声弹性成像通常只能得到组织在外部压缩作用下的纵向应变分布.超声弹性成像逆问题的求解,即重建组织内部的弹性模量分布,具有重要的意义.本文提出一种利用组织的应力-应变关系和基于有限元分析的迭代计算方法,从组织的纵向应变分布重建出组织的弹性模量分布.对平面应变状态下的均匀组织内含一圆形异物的模型和弹性模量连续过渡的模型分别进行了计算机仿真.结果初步验证了该方法的可行性.     

Multi-scale hierarchy of Chelydra serpentina: microstructure and mechanical properties of turtle shell

Carapace, the protective shell of a freshwater snapping turtle, Chelydra serpentina, shields them from ferocious attacks of their predators while maintaining light-weight and agility for a swim. The microstructure and mechanical properties of the turtle shell are very appealing to materials scientists and engineers for bio-mimicking, to obtain a multi-functional surface. In this study, we have elucidated the complex microstructure of a dry Chelydra serpentina’s shell which is very similar to a multi-layered composite structure. The microstructure of a turtle shell’s carapace elicits a sandwich structure of waxy top surface with a harder sub-surface layer serving as a shielding structure, followed by a lamellar carbonaceous layer serving as shock absorber, and the inner porous matrix serves as a load-bearing scaffold while acting as reservoir of retaining water and nutrients. The mechanical properties (elastic modulus and hardness) of various layers obtained via nanoindentation corroborate well with the functionality of each layer. Elastic modulus ranged between 0.47 and 22.15 GPa whereas hardness varied between 53.7 and 522.2 MPa depending on the microstructure of the carapace layer. Consequently, the modulus of each layer was represented into object oriented finite element (OOF2) modeling towards extracting the overall effective modulus of elasticity (∼4.75 GPa) of a turtle’s carapace. Stress distribution of complex layered structure was elicited with an applied strain of 1% in order to understand the load sharing of various composite layers in the turtle’s carapace.