基于深度学习的蝴蝶科级标本图像自动识别

【目的】本研究旨在探讨深度学习模型在蝴蝶科级标本图像自动识别中的可行性和泛化能力。【方法】为了提高识别模型的鲁棒性和泛化能力,将锤角亚目中6个科1 117种蝴蝶标本图像通过水平翻转、增加图像对比度与亮度以及添加噪声的方式增强图像数据集。在Caffe框架下,利用迁移学习方法,首先使用Image Net数据集中的图像训练Caffe Net模型,迭代31万次后得到初始化的网络权值;然后利用蝴蝶图像训练已预训练好的Caffe Net模型,通过参数微调,获得一个蝴蝶科级标本图像自动识别的卷积神经网络模型。为了比较深度学习和传统模式识别两种方法建立的模型的泛化能力,对相同训练样本提取全局特征和局部特征,训练支持向量机(support vector machine,SVM)分类器。所有的模型在与训练样本图像来源一致和不一致的两个测试样本集上进行测试。【结果】当测试样本与训练样本来源一致,均为蝴蝶标本图像时,基于Caffe Net的蝴蝶识别模型对6个科的蝴蝶识别准确率平均达到95.8%,基于Gabor的SVM分类器也获得了94.8%的识别率。当测试样本与训练样本来源不一致,为自然环境下拍摄的蝴蝶图像时,两种方法获得的识别率均下降,但Caffe Net模型对蝴蝶自然图像的平均识别率仍能达到65.6%,而基于Gabor的SVM分类器的识别率仅为38.9%。【结论】利用Caffe Net模型进行蝴蝶科级标本图像识别是可行的,相比较传统模式识别方法,基于深度学习的蝴蝶识别模型具有更好的泛化能力。     

实蝇科果实蝇属昆虫数字图像自动识别系统的构建和测试

针对双翅目实蝇科果实蝇属昆虫的自动识别,本文提出利用翅及中胸背板图像的局部二进制模式（local binary pattern, LBP）特征,采用Adaboost算法, 设计和开发“实蝇科果实蝇属昆虫数字图像自动识别系统”（Automated Fruit fly Identification System-Bactrocera, AFIS-B）。该系统包括图像采集、图像裁剪、预处理、特征提取、分类器设计、识别和显示,共7个模块。研究结果表明： LBP特征可以有效鉴别实蝇科果实蝇属昆虫;在对实蝇科果实蝇属8个种的测试中, 该系统表现出较高的准确性和稳定性,平均识别率可达80%以上。此外,还对果实蝇属昆虫翅膀及中胸背板图像在光照不均匀、姿态扭曲、样本受损及样本量大小等不同条件下的识别率进行了试验测试。结果表明, 该系统对测试样本的光照不均匀、 姿态扭曲和样本受损都表现出良好的鲁棒性,正确识别率与训练集样本各个种数量在一定条件下明显正相关,与训练集样本物种总量负相关。该项研究为实蝇科有害昆虫自动识别系统的构建及实际应用提供了理论、 方法及基础数据的支撑, 亦可为其他昆虫自动识别系统的研究和构建提供有益借鉴。 关键词：     

早期胚胎全场OCT图像的边界去模糊研究

全场光学相干层析成像技术(全场OCT)是研究早期胚胎形态发育的最理想成像设备,然而所采集图像难免受噪声干扰.这些噪声可模糊早期胚胎内不同组织结构的边界,从而给基于图像边界的结构划分带来干扰.为解决这一问题,本文运用中值滤波、维纳滤波、各向异性扩散算法处理全场OCT获得的早期胚胎图像,并运用信噪比、均方误差、峰值信噪比和边缘保留等指标评价图像处理效果.结果表明:经各向异性扩散算法处理的早期胚胎图像,可完整地保留原始图像信息,且边界最清晰,视觉效果最好.     

基于PET医学图像边缘检测算法的研究

目的:边缘检测在图像处理中至关重要,可被广泛应用于目标区域识别、区域形状检测、图像分割等图像分析领域。边缘是图像中不平稳现象和不规则结构的重要表现,往往携带着图像中的大量信息,并给出图像轮廓。在医学图像三维显示技术中,为了更精确的临床判别需要得到单像素的清晰轮廓,因此我们提出一种新的边缘检测算法。方法:在传统的小波边缘检测的基础上,提出了一种新的边缘算法,即基于小波极大值边缘检测算法,应用模糊算法构造相应的隶属函数,再对得到的极大值进一步筛选。结果:将该算法应用到医学图像中,最终可以得到较清楚的单像素边缘轮廓,实验结果证明了该算法的可行性。结论:运用这种算法处理过的医学图像边缘锐化更好,更清晰,能够为肿瘤的早期识别提供依据,满足医学影像识别的需要。     

虹膜识别中的一种神经算法的研究

研究了与传统身份识别不同的虹膜识别,在获取虹膜采样图后,经小波变换、松弛神经网络算法及稀疏编码,形成虹膜纹理的代码。并以海明距离作为决策空间的判决标准,实现人的虹膜图像识别的全过程,并对一些特定干扰条件下的虹膜图像进行计算机模拟试验,取得较好的结果。     

染色体易位重组位点的自动识别方法研究

染色体易位重组位点的识别对很多染色体遗传性疾病的诊断有着重要的意义。本文基于实际诊断中采集到的24类染色体数据和9号正常与异常染色体数据,构建了一套自动识别染色体易位重组位点的模型和方法。首先,对染色体图像进行预处理,得到了方向梯度直方图特征(HOG)和局部二值模式特征(LBP),构建了基于纹理特征的染色体24分类多通道网络模型,分类准确率达到了95.99%;再与ResNet18模型(分类准确率95.86%)进行模型融合,最终分类准确率达到97.08%。其次,将染色体密度谱作为正常和异常染色体的分类特征,采用投票的方法集成支持向量机、随机森林和XGBoost模型,构建了正常和异常染色体的集成分类器,正常和异常9号染色体的分类准确率达到了100%。最后,对于易位的异常染色体,我们提出了基于动态时间规划(DTW)的易位重组位点自动识别算法,在异常染色体的密度谱曲线上找到了重组位点,并映射至染色体G显带模式图,得到标准诊断结果,通过与临床专家的诊断结果进行比较说明了自动识别结果的有效性。本文设计的一套自动识别染色体易位重组位点的模型方法对临床辅助诊断有很大的帮助,有望完善成为一套软件系统应用于临床诊断,提升相关疾病的诊断效率和准确率。     

触珠蛋白电泳图谱的神经网络分型

１２０例人血清触珠蛋白的电泳图谱经数字化图像扫描,用神经网络进行分型。着重分析了影响带型识别的电泳因素。为提高网络的识别能力,将图谱按几种不同的方法进行数字化加工,得到了９４％的正确识别率。     

基于深度残差神经网络迁移学习的 牙形刺图像识别

为了探索基于深度神经网络模型的牙形刺图像智能识别效果, 研究选取奥陶纪8种牙形刺作为研究对象, 通过体视显微镜采集牙形刺图像1188幅, 收集整理公开发表文献的牙形刺图像778幅, 将图像数据集划分为训练集和测试集。通过对训练集图像进行旋转、翻转、滤波增强处理, 解决了训练样本不足的问题。基于ResNet-18、ResNet-34、ResNet-50、ResNet-101、ResNet-152五种残差神经网络模型, 采用迁移学习方法, 对网络模型进行训练以获取模型参数, 五种模型测试Top-1准确率分别为85.37%、85.85%、83.90%、81.95%、80.00%, Top-2准确率分别为94.63%、94.63%、94.15%、93.17%、93.66%, 模型对牙形刺图像具有较好的识别效果。通过对比研究发现, ResNet-34识别准确率最高, 说明对于特征简单的牙形刺属种, 增加网络深度并不一定能提升准确率, 而确定合适深度的模型则不仅可以提高识别准确率, 还可以节约计算资源。通过ResNet-34模型的迁移学习训练和重新训练效果对比可以看出, 迁移学习不仅可以获得较高的准确率, 而且可以较快获取模型参数, 因而可作为小样本古生物化石图像识别的重要方法。研究还发现, 体视显微镜下牙形刺图像的识别准确率高于扫描电镜下图像识别准确率, 化石完整性和相似性、照相角度以及数据集的大小是影响图像识别准确率的主要原因。     

基于颜色直方图及双树复小波变换（DTCWT）的昆虫图像识别

为了给生产单位害虫管理的普通技术人员提供简便易操作的昆虫鉴别方法, 本文提出了一种新颖的基于图像颜色及纹理特征的昆虫图像识别方法。鳞翅目昆虫翅面图像经过预处理, 确定目标区域, 再进行特征提取。首先将彩色图像从三原色(red-green-blue, RGB)空间转换至色调饱和值（HSV）空间并提取有效区域内的色度、饱和度直方图特征, 然后经图像位置校准, 提取灰度图的双树复小波变换（DTCWT）特征; 匹配首先计算两颜色直方图特征向量之间的相关性, 将相关性大于阈值的样本再进一步用DTCWT特征匹配; DTCWT匹配通过计算Canberra距离实现, 从通过第一层颜色匹配的样本中取出最近邻作为最终匹配类别。算法在包含100类鳞翅目昆虫的图像库中进行试验验证, 取得了76%的识别率, 其中前翅识别率则达92%, 同时取得了理想的时间性能。试验结果证明了本文方法的有效性。     

基于K-means聚类算法的叶螨图像分割与识别

【目的】叶螨(spider mite)是为害多种农作物的主要害虫,叶螨识别传统方法依靠肉眼,比较费时费力,为研究快速自动识别方法,引入计算机图像分析算法。【方法】该方法基于K-means聚类算法对田间作物上的叶螨图像进行分割与识别。【结果】对比传统RGB彩色分割方法,K-means聚类算法能够有效地对叶片上叶螨图像进行分割和识别。K-means聚类算法平均识别时间为3.56 s,平均识别准确率93.95%。识别时间 T 随图像总像素 Pi 的增加而增加。【结论】K-means聚类组合算法能够应用于叶螨图像分割与识别。     

Imaging performance of phase-contrast breast computed tomography with synchrotron radiation and a CdTe photon-counting detector

PurposeWithin the SYRMA-CT collaboration based at the ELETTRA synchrotron radiation (SR) facility the authors investigated the imaging performance of the phase-contrast computed tomography (CT) system dedicated to monochromatic in vivo 3D imaging of the female breast, for breast cancer diagnosis.MethodsTest objects were imaged at 38 keV using monochromatic SR and a high-resolution CdTe photon-counting detector. Signal and noise performance were evaluated using modulation transfer function (MTF) and noise power spectrum. The analysis was performed on the images obtained with the application of a phase retrieval algorithm as well as on those obtained without phase retrieval. The contrast to noise ratio (CNR) and the capability of detecting test microcalcification clusters and soft masses were investigated.ResultsFor a voxel size of (60 μm)³, images without phase retrieval showed higher spatial resolution (6.7 mm⁻¹ at 10% MTF) than corresponding images with phase retrieval (2.5 mm⁻¹). Phase retrieval produced a reduction of the noise level and an increase of the CNR by more than one order of magnitude, compared to raw phase-contrast images. Microcalcifications with a diameter down to 130 μm could be detected in both types of images.ConclusionsThe investigation on test objects indicates that breast CT with a monochromatic SR source is technically feasible in terms of spatial resolution, image noise and contrast, for in vivo 3D imaging with a dose comparable to that of two-view mammography. Images obtained with the phase retrieval algorithm showed the best performance in the trade-off between spatial resolution and image noise.     

Diagnostic Accuracy of Lipopolysaccharide-Binding Protein as Biomarker for Sepsis in Adult Patients: A Systematic Review and Meta-Analysis

IntroductionLipopolysaccharide-binding protein (LBP) is widely reported as a biomarker to differentiate infected from non-infected patients. The diagnostic use of LBP for sepsis remains a matter of debate. We aimed to perform a systematic review and meta-analysis to assess the diagnostic accuracy of serum LBP for sepsis in adult patients.MethodsWe performed a systematic review and meta-analysis to assess the accuracy of LBP for sepsis diagnosis. A systematic search in PubMed and EMBASE for studies that evaluated the diagnostic role of LBP for sepsis through December 2015 was conducted. We searched these databases for original, English language, research articles that studied the diagnostic accuracy between septic and non-septic adult patients. Sensitivity, specificity, and other measures of accuracy, such as diagnostic odds ratio (DOR) and area under the receiver operating characteristic curve (AUC) of LBP were pooled using the Hierarchical Summary Receiver Operating Characteristic (HSROC) method.ResultsOur search returned 53 reports, of which 8 fulfilled the inclusion criteria, accounting for 1684 patients. The pooled sensitivity and specificity of LBP for diagnosis of sepsis by the HSROC method were 0.64 (95% CI: 0.56–0.72) and 0.63 (95% CI: 0.53–0.73), respectively. The value of the DOR was 3.0 (95% CI: 2.0–4.0) and the AUC was 0.68 (95% CI: 0.64–0.72). Meta-regression analysis revealed that cut-off values accounted for the heterogeneity of sensitivity and sample size (> = 150) accounted for the heterogeneity of specificity.ConclusionsBased on the results of our meta-analysis, LBP had weak sensitivity and specificity in the detection of sepsis. LBP may not be practically recommended for clinical utilization as a single biomarker.     

Can the non-pre-whitening model observer,including aspects of the human visual system,predict human observer performance in mammography?

PurposeIn mammography, images are processed prior to display. Current methodologies based on physical image quality measurements are however not designed for the evaluation of processed images. Model observers (MO) might be suitable for this evaluation. The aim of this study was to investigate whether the non-pre-whitening (NPW) MO can be used to predict human observer performance in mammography-like images by including different aspects of the human visual system (HVS).MethodsThe correlation between human and NPW MO performance has been investigated for the detection of disk shaped objects in simulated white noise (WN) and clustered lumpy backgrounds (CLB), representing quantum noise limited and mammography-like images respectively. The images were scored by the MO and five human observers in a 2-alternative forced choice experiment.ResultsFor WN images it was found that the log likelihood ratio (R_LR²), which expresses the goodness of fit, was highest (0.44) for the NPW MO without addition of HVS aspects. For CLB the R_LR² improved from 0.46 to 0.65 with addition of HVS aspects. The correlation was affected by object size and background.ConclusionsThis study shows that by including aspects of the HVS, the performance of the NPW MO can be improved to better predict human observer performance. This demonstrates that the NPW MO has potential for image quality assessment. However, due to the dependencies found in the correlation, the NPW MO can only be used for image quality assessment for a limited range of object sizes and background variability.     

Comparison between new-generation SiPM-based and conventional PMT-based TOF-PET/CT

PurposeThis study aimed to determine whether the SiPM-PET/CT, Discovery MI (DMI) performs better than the PMT-PET/CT system, Discovery 710 (D710).MethodsThe physical performance of both systems was evaluated using NEMA NU 2 standards. Contrast (%), uniformity and image noise (%) are criteria proposed by the Japanese Society of Nuclear Medicine (JSNM) for phantom tests and were determined in images acquired from Hoffman and uniform phantoms using the DMI and D710. Brain and whole-body [¹⁸F]FDG images were also acquired from a healthy male using the DMI and D710.ResultsThe spatial resolution at 1.0 cm off-center in the DMI and D710 was 3.91 and 4.52 mm, respectively. The sensitivity of the DMI and D710 was 12.62 and 7.50 cps/kBq, respectively. The observed peak noise-equivalent count rates were 185.6 kcps at 22.5 kBq/mL and 137.0 kcps at 29.0 kBq/mL, and the scatter fractions were 42.1% and 37.9% in the DMI and D710, respectively. The D710 had better contrast recovery and lower background variability. Contrast, uniformity and image noise in the DMI were 61.0%, 0.0225, and 7.85%, respectively. These outcomes were better than those derived from the D710 and satisfied the JSNM criteria. Brain images acquired by the DMI had better grey-to-white matter contrast and lower image noise at the edge of axial field of view.ConclusionsThe DMI offers better sensitivity, performance under conditions of high count rates and image quality than the conventional PMT-PET/CT system, D710.     

Technical Note: Quality assessment of virtual monochromatic spectral images on a dual energy CT scanner

PurposeTo assess the quality of images obtained on a dual energy computed tomography (CT) scanner.MethodsImage quality was assessed on a 64 detector-row fast kVp-switching dual energy CT scanner (Revolution GSI, GE Medical Systems). The Catphan phantom and a low contrast resolution phantom were employed. Acquisitions were performed at eight different radiation dose levels that ranged from 9 mGy to 32 mGy. Virtual monochromatic spectral images (VMI) were reconstructed in the 40–140 keV range using all available kernels and iterative reconstruction (IR) at four different blending levels. Modulation Transfer Function (MTF) curves, image noise, image contrast, noise power spectrum and contrast to noise ratio were assessed.ResultsIn-plane spatial resolution at the 10% of the MTF curve was 0.60 mm⁻¹. In-plane spatial resolution was not modified with VMI energy and IR blending level. Image noise was reduced from 16.6 at 9 mGy to 6.7 at 32 mGy, while peak frequency remained within 0.14 ± 0.01 mm⁻¹. Image noise was reduced from 14.3 at IR 10% to 11.5 at IR 50% at a constant peak frequency. The lowest image noise and maximum peak frequency were recorded at 70 keV.ConclusionsOur results have shown how objective image quality is varied when different levels of radiation dose and different settings in IR are applied. These results provide CT operators an in depth understanding of the imaging performance characteristics in dual energy CT.     

Diagnosing Discogenic Low Back Pain Associated with Degenerative Disc Disease Using a Medical Interview

PurposesTo evaluate the usefulness of our original five questions in a medical interview for diagnosing discogenic low back pain (LBP), and to establish a support tool for diagnosing discogenic LBP.ResultsThere were no significant differences in baseline characteristics, except age, between the groups. There were significant differences between the groups for all five questions. In the age-adjusted analyses, the odds ratios of LBP after sitting too long, while standing after sitting too long, squirming in a chair after sitting too long, while washing one’s face, and in standing position with flexion were 10.5, 8.5, 4.0, 10.8, and 11.8, respectively. The integer scores were 11, 9, 4, 11, and 12, respectively, and the sum of the points of the five scores ranged from 0 to 47. Results of the ROC analysis were as follows: cut-off value, 31 points; area under the curve, 0.92302; sensitivity, 100%; and specificity, 71.4%.ConclusionsAll five questions were useful for diagnosing discogenic LBP. We established the scoring system as a support tool for diagnosing discogenic LBP.     

Optimized Two-Stage Ensemble Model for Mammography Mass Recognition

ObjectivesMammography mass recognition is considered as a very challenge pattern recognition problem due to the high similarity between normal and abnormal masses. Therefore, the main objective of this study is to develop an efficient and optimized two-stage recognition model to tackle this recognition task.Material and methodsBasically, the developed recognition model combines an ensemble of linear Support Vector Machine (SVM) classifiers with a Reinforcement Learning-based Memetic Particle Swarm Optimizer (RLMPSO) as RLMPSO-SVM recognition model. RLMPSO is used to construct a two-stage of an ensemble of linear SVM classifiers by performing simultaneous SVM parameters tuning, features selection, and training instances selection. The first stage of RLMPSO-SVM recognition model is responsible about recognizing the input ROI mammography masses as normal or abnormal mass pattern. Meanwhile, the second stage of RLMPSO-SVM model used to perform further recognition for abnormal ROIs as malignant or benign masses. In order to evaluate the effectiveness of RLMPSO-SVM, a total of 1187 normal ROIs, 111 malignant ROIs, and 135 benign ROIs were randomly selected from DDSM database images.ResultsReported results indicated that RLMPSO-SVM model was able to achieve performances of 97.57% sensitivity rate with 97.86% specificity rate for normal vs. abnormal recognition cases. For malignant vs. benign recognition performance it was reported of 97.81% sensitivity rate with 96.92% specificity rate.ConclusionReported results indicated that RLMPSO-SVM recognition model is an effective tool that could assist the radiologist during the diagnosis of the presented abnormalities in mammography images. The outcomes indicated that RLMPSO-SVM significantly outperformed various SVM-based models as well as other variants of computational intelligence models including multi-layer perceptron, naive Bayes classifier, and k-nearest neighbor.     

Quercetin intake during lactation modulates the AMP-activated protein kinase pathway in the livers of adult male rat offspring programmed by maternal protein restriction

Quercetin, a naturally occurring flavonoid, has been reported to possess numerous biological activities including activation of adenosine-5’-monophosphate-activated protein kinase (AMPK). We investigated the effects of quercetin intake during lactation on the AMPK activation in the livers of adult offspring programmed by maternal protein restriction during gestation. Pregnant Wistar rats were fed control and low-protein diets during gestation. Following delivery, each dam received a control or 0.2% quercetin-containing control diet during lactation as follows: control on control (CC), control on restricted (LPC) and 0.2% quercetin-containing control on restricted (LPQ). At weaning (week 3), some of the pups from each dam were killed, and the remaining pups (CC, n= 8; LPC, n= 10; LPQ, n= 13) continued to receive a standard laboratory diet and were killed at week 23. Blood chemistry and phosphorylation levels of AMPKα, acetyl-CoA carboxylase (ACC), endothelial nitric oxide synthase (eNOS) and mammalian target of rapamycin (mTOR) in the livers of male offspring were examined. At week 3, the level of phosphorylated AMPK protein in LPQ increased about 1.5- and 2.1-fold compared with LPC and CC, respectively, and the level in LPQ at week 23 increased about 1.9- and 2.9-fold, respectively. A significant increase in phosphorylated ACC and eNOS levels was found in LPQ. There was no significant difference among the three groups in the level of phosphorylated mTOR protein. In conclusion, quercetin intake during lactation up-regulates AMPK activation in the adult offspring of protein-restricted dams and modulates the AMPK pathway in the liver.     

一种非酒精性脂肪性肝病的超声影像人工智能辅助诊断系统开发

摘要 目的：结合人工智能方法设计针对肝脏超声影像的辅助诊断系统，辅助医生对大样本肝脏超声影像数据的标准化和高效化诊断，实现基于肝脏超声图像的非酒精性脂肪性肝病的精准诊断。方法：通过开发肝脏超声影像的识别与分类、脂肪肝分级分析和肝脏脂肪含量定量分析三个模块，建立一套非酒精性脂肪性肝病的超声影像人工智能辅助诊断系统，该系统能够自动区分输入到系统中不同采样视野的超声影像类型，并对肝脏超声图像进行数字化分析，给出待测超声图像是否呈现脂肪肝以及其肝脏脂肪含量的百分比值。结果：本研究中的超声图像识别分类模块可高通量区分出肝肾比图像和衰减率图像的两类超声影像，其分类的准确率达100%。脂肪肝分级分析模块在测试集数据的准确率达到84%，展现出可胜任辅助医生诊断的能力。基于人工肝脏脂肪含量定量方法开发的肝脏脂肪含量定量分析模块的准确率达到67.74%。结论：本研究已开发出一套基于肝脏超声影像的智能辅助诊断系统，可以辅助医生快速、简单、无创地筛选出潜在患有脂肪肝的患者，虽然现阶段实现肝脏脂肪定量分析仍有难度，但已展现出较大的临床应用潜力。