3D打印技术在股骨远端骨肿瘤的应用

目的：寻找适合识别正常肝、原发性肝癌和肝血管瘤CT图像的特征向量。方法：从一阶统计特征、灰度共生矩阵、灰度行程矩阵三方面提取正常肝、原发性肝癌和肝血管瘤CT图像的纹理特征,然后采用t检验进行特征选择,最后利用神经网络识别系统,把保留的特征作为输入量,对正常肝和原发性肝癌进行识别。结果：所设计的BP神经网络,对正常肝（100±0．00）％、原发性肝癌（91．08±6．96）％,对肝血管瘤（85．76±12．51）％。结论：BP神经网络经设计优化后能达到较高的识别准确率,对于原发性肝癌的计算机辅助诊断具有一定的实际意义和理论价值。     

Automated Identification of Diabetic Retinopathy Stages Using Digital Fundus Images

Diabetic retinopathy (DR) is caused by damage to the small blood vessels of the retina in the posterior part of the eye of the diabetic patient. The main stages of diabetic retinopathy are non-proliferate diabetes retinopathy (NPDR) and proliferate diabetes retinopathy (PDR). The retinal fundus photographs are widely used in the diagnosis and treatment of various eye diseases in clinics. It is also one of the main resources for mass screening of diabetic retinopathy. In this work, we have proposed a computer-based approach for the detection of diabetic retinopathy stage using fundus images. Image preprocessing, morphological processing techniques and texture analysis methods are applied on the fundus images to detect the features such as area of hard exudates, area of the blood vessels and the contrast. Our protocol uses total of 140 subjects consisting of two stages of DR and normal. Our extracted features are statistically significant (p < 0.0001) with distinct mean ± SD as shown in Table 1. These features are then used as an input to the artificial neural network (ANN) for an automatic classification. The detection results are validated by comparing it with expert ophthalmologists. We demonstrated a classification accuracy of 93%, sensitivity of 90% and specificity of 100%.     

基于U型全卷积神经网络的腹部动脉CT图像分割算法研究

目的为了精确分割腹部动脉血管,提出一种基于深度学习的全自动腹部动脉CT图像分割算法。方法采用区域不平衡块生成方法提取CT血管横断面、冠状面和矢状面图像特征,接着采用U型全卷积神经网络对块特征进行训练与分割,最后采用最大体素保留法获得三维血管分割图像。选用120例患者腹部CT血管图像进行网络训练和分割实验,分割结果评价指标采用精确率、召回率和Dice系数。结果基于U型全卷积神经网络能分割全部腹部CT图像大血管和绝大多数小血管。全卷积神经网络中块尺寸s=32所得平均Dice系数、精确率和召回率分别达87.2%、85.9%和88.5%,且与块尺寸s=48和s=64大致相等。基于U型全卷积神经网络所得平均Dice系数、精确率和召回率均优于其他血管分割算法。结论基于U型全卷积神经网络算法的图像分割精度高,是一种可行的腹部CT血管分割算法。     

Automated Analysis of Retinal Vascular Tortuosity on Color Retinal Images

Recent advances in medical imaging modality have enabled us to identify new features in retinal vasculature. One of the features is retinal vascular tortuosity which has been shown to become a predictive factor for cardiovascular diseases and diabetes. The changes in retinal vascular tortuosity might be a sign of severity or improvement of the disease. In this paper, we propose a new method for measuring retinal vascular tortuosity. We adopt a new technique to analyze tortuosity that consider vessel-segment’s width simultaneously. Our proposed method measures vessel-segment’s tortuosity on its edge. A qualitative assessment shows that the method is appropriate for measuring the tortuosity of the vessels in different widths and directions in the image. Finally, a comparison distinguishing tortuous vs. non tortuous vessels demonstrates that the proposed approach may be suitable for predicting or earlier diagnosis of diabetes or cardiovascular diseases.     

基于支持向量机的骨肉瘤X线图像中病变区域的自动识别研究

采用骨肉瘤X线图像中病变区域的颜色特征和基于灰度共生矩阵的纹理特征作为特征向量,研究利用支持向量机算法对骨肉瘤病变区域的自动识别方法,结果充分表明支持向量机良好的分类能力.     

Highly Sensitive Computer Aided Diagnosis System for Breast Tumor Based on Color Doppler Flow Images

A computer-aided diagnosis (CAD) system for breast tumor based on color Doppler flow images is proposed. Our system consists of automatic segmentation, feature extraction, and classification of breast tumors. First, the B-mode grayscale image containing anatomical information was separated from a color Doppler flow image (CDFI). Second, the boundary of the breast tumor was automatically defined in the B-mode image and then morphologic and gray features were extracted. Third, an optimal feature vector was created using K-means cluster algorithm. Then a back-propagation (BP) artificial neural network (ANN) was used to classify breast tumors as benign, malignant or uncertain. Finally, the blood flow feature was extracted selectively from the CDFI, and was used to classify the uncertain tumor as benign or malignant. Experiments on 500 cases show that the proposed system yields an accuracy of 100% for the malignant and 80.8% for the benign classification. Comparing with other systems, the advantage of our system is that it has a much lower percentage of malignant tumor misdiagnosis.     

基于机器学习的椎弓根螺钉钉道超声图像评估方法

目的 探讨基于支持向量机（SVM）构建的人工智能辅助诊断模型对椎弓根螺钉钉道完整性进行超声鉴别与验证的方法研究。 方法 本文提出了一种基于超声图像智能分析的椎弓根钉道完整性评估方法。数据采自4个新鲜人体胸腰椎标本。预建立钉道50个，共800张超声图像（其中钉道完整与破损的样本各400个），采用五折交叉验证的方法对样本进行训练集与测试集的划分，对人工智能辅助诊断模型进行训练及测试。首先对超声图像进行裁剪，并采用亮度映射方法进行图像增强得到易于计算机判断识别且排除无效信息干扰的超声图像；然后通过灰度共生矩阵算法进行纹理特征提取，并采用支持向量机模型对正常和严重破损样本的初始分类模型进行搭建；其次，使用灰度分布得到用于区分前景和背景的阈值，并通过设计的损失函数得到得到钉道同心圆的半径；最后将同心圆外部图像的熵、方差、对比度、能量、平均绝对偏差作为第二类特征，最后进行轻微破损样本和未破损样本的二次分类模型搭建。 结果 初始分类的准确率为74.75%，特异性为71.81%，灵敏度为81.5%，F1值为76.35%，假正率为32%，假负率为18.5%。二次分类的准确率为93.75%，特异性为91.55%，灵敏度为97.5%，F1值为94.43%，假正率为9%，假负率为2.5%。二次类准确率与初始分类相比较，准确率提升19%。 结论 本文提出的基于SVM机器学习模型的方法可较为准确地检测椎弓根钉道的破损情况，且准确率较高，可用于术中实时判断椎弓根钉道的状态。     

基于迁移学习的全连接神经网络舌象分类方法

目的 针对深度学习在舌象分类中训练数据量大、训练设备要求高、训练时间长等问题,提出一种基于迁移学习的全连接神经网络小样本舌象分类方法。方法 应用经ImageNet海量数据集训练后的卷积Inception_v3网络提取舌象点、线等有效特征,再使用全连接神经网络对特征进行训练分类,将深度学习网络学习到的图像知识迁移到舌象识别任务中。利用舌象数据集进行训练、测试。结果 与典型舌象分类方法K最近邻（KNN）算法、支持向量机（SVM）算法和卷积神经网络（CNN）深度学习方法相比,本实验使用的两种方法（Inception_v3+2NN和Inception_v3+3NN）具有较高的舌象分类识别率,准确率分别达90.30%和93.98%,且样本训练时间明显缩短。结论 与KNN算法、SVM算法和CNN深度学习方法相比,基于迁移学习的全连接神经网络舌象分类方法可有效提高舌象分类的准确率、缩短网络的训练时间。     

一种基于BP网络的CT图像肝实质分割算法

提出一种基于BP网络分割CT图像序列中肝实质的方法。首先选取训练样本，提取样本图像中肝脏的纹理特征，作为输入向量，以对训练样本手工分割的结果作为导师信号，对BP神经网络进行训练，再用训练好的网络对CT图像序列中的肝实质进行分割，最后对分割后的结果进行三维区域生长及孔洞填充处理。实验结果表明：该方法能够有效的对肝脏纹理特征明显的CT图像序列进行分割，可用于CT图像序列的自动分割。     

基于多尺度小波残差网络的稀疏角度CT图像恢复

目的 稀疏角度CT具有加速数据采集和减少辐射剂量的优点。然而,由于采集信息的减少,使用传统滤波反投影算法（FBP）进行重建得到的图像中伴有严重的条形伪影和噪声。针对这一问题,本文提出基于多尺度小波残差网络（MWResNet）对稀疏角度CT图像进行恢复。方法 本网络中将小波网络与残差块相结合,用以增强网络对图像特征的提取能力和加快网络训练效率。实验中使用真实的螺旋几何CT图像数据“Low-dose CT Grand Challenge”数据集训练网络。通过观察图像表征和计算定量参数的方法对结果进行评估,并与其他现有网络进行比较,包括图像恢复迭代残差卷积网络（IRLNet）,残差编码解码卷积神经网络（REDCNN）和FBP卷积神经网络（FBPConvNet）。结果 实验结果表明,本文提出的多尺度小波残差网络优于其余对比方法。结论 本文提出的MWResNet网络能够在保持稀疏角度CT图像边缘细节信息的同时有效抑制噪声和伪影。     

Development of a Robust Algorithm for Detection of Nuclei and Classification of White Blood Cells in Peripheral Blood Smear Images

Peripheral Blood Smear analysis plays a vital role in diagnosis of many diseases such as leukemia, anemia, malaria, lymphoma and infections. Unusual variations in color, shape and size of blood cells indicate abnormal condition. We used a total of 117 images from Leishman stained peripheral blood smears acquired at a magnification of 100X. In this paper we present a robust image processing algorithm for detection of nuclei and classification of white blood cells based on features of the nuclei. We used novel image enhancement method to manage illumination variations and TissueQuant method to manage color variations for the detection of nuclei. Dice similarity coefficient of 0.95 was obtained for nucleus detection. We also compared the proposed method with a state-of-the-art method and the proposed method was found to be better. Shape and texture features of the detected nuclei were used for classifying white blood cells. We considered classification of WBCs using two approaches such as 5-class and cell-by-cell approaches using neural network and hybrid-classifier respectively. We compared the results of both the approaches for classification of white blood cells. Cell-by-cell approach offered 1.4% higher sensitivity in comparison with the 5-class approach. We obtained an accuracy of 100% for lymphocyte and basophil detection. Hence, we conclude that lymphocytes and basophils can be accurately detected even when the analysis is limited to the features of nuclei whereas, accurate detection of other types of WBCs will require analysis of the cytoplasm too.     

手掌静脉识别：基于端到端卷积神经网络方法

目的提出一种基于端到端卷积神经网络的手掌静脉识别方法。方法在构建的手掌静脉识别网络模型中,卷积层和池化 层交替级联提取图像特征,同时通过神经网络分类器进行分类识别,采用包含动量项的随机梯度下降法最小化识别误差,在误 差减小的方向上不断优化模型。采用训练集数据扩展、批归一化、Dropout、L2参数正则化四种方法提升网络的泛化能力。结果 对公共的PolyU库(图像在高约束条件下获取)和自建库（图像在自然条件下获取）中全部500个对象的识别,正确识别率分别达 到99.90%和98.05%,单个样本的识别时间均小于9 ms。结论与传统算法相比,本文方法能够有效提升掌静脉识别在实际应用 中的准确率,为掌静脉识别提供一种新思路。     

Lung Cancer Detection Using Fuzzy Auto-Seed Cluster Means Morphological Segmentation and SVM Classifier

An effective fuzzy auto-seed cluster means morphological algorithm developed in this work to segment the lung nodules from the consecutive slices of Computer Tomography (CT) images to detect the lung cancer. The initial cluster values were chosen automatically by averaging the minimum and maximum pixel values in each row of an image. The area and eccentricity features were used to eliminate the line like structure and very tiny clusters less than 3 mm in size. The change in centroid analysis was carried out to eliminate the blood vessels. The tissue clusters whose centroid varies much in consecutive slices must be blood vessels. After eliminating the blood vessels, the co-occurrence matrix based texture features contrast, homogeneity and auto correlation were computed on the remaining nodules from the consecutive CT slices to discriminate the calcifications. The extracted centroid shift and texture features were used as the inputs to the Support Vector Machine (SVM) kernel classifier in order to classify the real malignant nodules. This work was carried out on 56 malignant (cancerous) cases and 50 normal cases (with lung infections), which had a total of 56 malignant nodules and 745 benign nodules. Out of these, 60 % of subjects (34 cancerous & 30 non-cancerous) were used for training. The remaining 40 % subjects (22 cancerous & 20 non-cancerous) were used for testing. This work produced a good sensitivity, specificity and accuracy of 100 %, 93 % and 94 %, respectively. The False Positive (FP) per patient was calculated as 0.38.     

国外医学伦理学合著网络分析

利用社会网络分析方法研究国外医学伦理学领域的作者合著情况,以《期刊引文报告》收录的国外医学伦理学9种核心期刊为数据源,通过BICOMB、UCINET软件抽取整理信息,生成作者共现矩阵并绘制图谱,揭示国外医学伦理学领域作者合著网络的整体结构特性和作者合作情况。     

科技论文网络研究进展及建立论文相似网络的构想

从复杂网络角度介绍科技论文相关网络的研究进展,包括引用网络和共现网络,在此基础上提出了建立论文内容相似性网络的构想,并探讨其现实意义与关键问题,包括论文内容的完整准确表达、利用向量空间模型表达文献内容、被引强度和耦合率等.     

基于形态学特征提取的人工神经网络在口腔鳞状细胞癌诊断中的应用

目的 基于形态学图像处理方法,应用径向基神经网络(radial basis function,RBF)寻找一种可行、便捷的方法辅助口腔鳞状细胞癌的诊断.方法 选择口腔鳞状细胞癌和口腔非癌的组织病理切片图像进行形态学方法处理,提取表述特征的向量,作为训练集训练RBF网络;另选择67帧病理图像,包含癌和非癌的病例,作为测试集观察RBF的性能.结果 在RBF网络将测试标本分类结果的分析中可以看到不同输出值分类阈值的选择对应不同的诊断敏感度和特异度.结论 训练后的RBF虽然鉴别阳性、阴性的能力不能和金标准(即病理诊断)相比,但是通过选择不同敏感度和特异度,依然能够有效辅助病理医师,提高诊断效率,发挥机器的优势.     

基于多尺度三维卷积神经网络的头颈部危及器官分割方法

目的 研究一种基于三维卷积神经网络的CT图像头颈部危及器官分割算法。方法 本文构建了一个基于V-Net模型的头颈部危及器官自动分割算法。为了增强分割模型的特征表达能力,将SE（Squeeze-and-Excitation）模块与V-Net模型中残差卷 积模块相结合,提高与分割任务相关性更大的特征通道权重;采用多尺度策略,使用粗定位和精分割两个级联模型完成器官分割,其中输入图像在预处理时重采样为不同分辨率,使得模型分别专注于全局位置信息和局部细节特征的提取。结果 我们在头颈部22个危及器官的分割实验表明,相比于已有方法,本文提出的方法分割平均精度提升了9%,同时平均测试时间从33.82 s降低至2.79 s。结论 基于多尺度策略的三维卷积神经网络达到了更好的分割精度,且耗时极短,有望在临床应用中提高医生的工作效率。     

基于ResNet深度网络的人类蛋白质图谱图像分类方法研究

将基于深度学习的图像分类方法引入人类蛋白质图谱图像分类中，利用ResNet深度网络构建面向人类蛋白质图谱图像分类的深度卷积神经网络，通过混合模式的蛋白质显微镜图像进行验证。结果表明该方法比其他自动分类法具有更高的准确率和精度，大大节约人力和时间。