基于多特征条件随机场的《金匮要略》症状药物信息抽取研究

目的结合自然语言处理方法,研究可以有效抽取中医古籍中所含症状和药物文本实体信息的方法。方法以《金匮要略》为例,采用条件随机场(CRF)算法,先将文本进行分词处理,然后以词性、基于键值对的中医诊断标记集作为辅助特征,通过症状-药物BIO标签为训练特征来训练出模型,然后利用该模型对测试集文本进行自动标签标注。结果基于多特征CRF自动标注的结果准确率达到84.5%,召回率达到70.9%,F测度值达到77.1%。结论运用CRF方法加入词性、中医诊断标记集特征集进行训练得出的多特征模型,能有效提高CRF算法对中医古籍的实体抽取能力,生成的模型可用来自动化抽取中医古籍文本的症状药物实体信息。     

基于自然语言处理的临床合理用药知识图谱构建

目的：构建基于自然语言处理的临床合理用药知识图谱。方法：以国家食品药品监督管理总局（CFDA）、美国食品药品监督管理总局（FDA）及某大型三甲医院药品库中药品说明书为数据源，构建了一种基于深度学习算法的临床合理用药知识图谱库。对随机抽取的500份药品说明书进行人工标注，将标注的数据划分为训练集、测试集、验证集。基于深度学习模型BRET进行训练，通过训练集训练模型和验证集验证训练过程中的性能及训练后通过测试集测试模型性能，用优化后的机器学习模型预测未标注的药品说明书。结果：最终抽取出30余万条“实体-关系-实体”的三元组关系，将机器学习模型产生的三元组与领域专家标注产生的三元组一起导入Neo4j图形数据库中存储，以知识图谱的形式展现给临床药师。结论：通过基于深度学习算法的临床合理用药知识库构建，在标引少量药品说明书的前提下，挖掘出药品说明书中所有的医疗关系和实体。自动构建基于药品说明书的合理用药知识图谱，可提高合理用药的自动化程度和准确度，降低不合理用药。     

基于医学大数据的预训练语言模型及其医学文本分类研究

目的:构建基于医学文本的预训练语言模型,以解决基于通用语料的预训练语言模型不适应医学文本分类的问题。方法:使用PubMed医学论文摘要数据和PMC医学论文全文数据在通用预训练语言模型Bert上进行二次预训练,得到医学领域的预训练语言模型BioBert,使用标注好的文本数据对BioBert进行微调,得到最终的医学文本分类模型。结果:病历文本和医学论文摘要文本两个数据集的分类实验显示,经过医学文本二次预训练的预训练语言模型在两个数据集上都取得了较好的分类效果。结论:通过自训练的方式对大量医学文本进行预训练得到的医学领域预训练语言模型,能在一定程度上解决使用通用预训练语言模型无法很好适配医学文本分布而导致分类性能偏低的问题。     

基于Wasserstein Gan的无监督单模配准方法

本文提出一种基于Wasserstein Gan的无监督单模配准方法。与现有的基于深度学习的单模配准方法不同，本文的方法完成训练不需要Ground truth和预设的相似性度量指标。本文方法的主要结构包括生成网络和判别网络。首先，生成网络输入固定图像（正例图像）和浮动图像并提取图像间潜在的形变场，通过插值方式预测配准图像（负例图像）；然后，判别网络交替输入正例图像和负例图像，判断图像间的相似性，并将判断结果作为损失函数反馈，进而驱动网络参数更新；最后，通过对抗训练，生成网络预测的配准图像能欺骗判别网络，网络收敛。实验中随机选取30例LPBA40脑部数据集、25例EMPIRE10肺部数据集和15例ACDC心脏数据集用作训练数据集，然后将剩下的10例LPBA40脑部数据集、5例EMPIRE10肺部数据集和5例ACDC心脏数据集用作测试数据集。配准结果与Affine算法、Demons算法、SyN算法和VoxelMorph算法对比。实验结果显示，本研究算法的DICE系数（DSC）和归一化相关系数（NCC）评价指标均是最高，表明本文方法的配准精度高于Affine算法、Demons算法、SyN算法和目前无监督的SOTA算法VoxelMorph。     

基于临床病历预训练语言模型的病历文本纠错探索

电子病历文本中存在错别字既不符合国家电子病历管理规范，又降低了自然语言处理技术的效果，影响了电子病历的价值挖掘与应用。阐述了一种基于在大量真实病历语料上训练出的预训练语言模型进行自动纠错的方法。实验证明，该方法在仿真数据集和真实病历数据集上检错和纠错都取得了很好的效果，运行效率很高，可以支持事中和事后的电子病历纠错，有效提升电子病历质量，推动电子病历的应用。     

中医药文献语义标注系统研究与开发

目的研究和开发支持中医和现代生物医学本体和术语集的语义标注系统。方法以MedPortal本体库和中医临床术语集等为术语资源库,设计语义标注系统工作流程和功能框架,并开发Web应用系统。结果构建了一个基于Web的中医药文献语义标注系统,支持语料库管理与维护、术语词典管理、语义标注和语义检索等功能,既可以为基于机器学习的信息抽取算法研究提供训练集,又能实现语义层面的多来源数据集成与知识融合。结论该中医药文献语义标注系统设计方案已经过实际项目验证,可为其他同类系统研发提供参考。     

基于深度学习的中孕期胎儿超声筛查切面自动识别

目的本研究利用深度学习方法,基于深度卷积神经网络模型,对中孕期胎儿超声筛查图像的31个标准切面进行自动识别。方法采集孕20~24周胎儿超声筛查切面图像共76260张(包含31个切面),将其划分为训练集68386张,测试集7874张。在Vgg16网络模型上进行模型微调,加载数据集进行训练。将训练好的模型在测试集进行验证。结果该模型对于胎儿超声筛查切面的识别正确率为94.8%。结论该方法能够准确识别胎儿超声筛查图像的每个切面,为胎儿超声图像的自动质量控制解决方案打下了坚实的基础。     

深度学习算法辅助构建三维颜面正中矢状平面

目的: 旨在建立一种可准确确定三维颜面解剖标志点的深度学习算法——多视图堆叠沙漏神经网络(multi-view stacked hourglass convolutional neural networks,MSH-CNN), 并结合赋权普氏分析算法实现三维颜面正中矢状平面的自动构建。方法: 收集面部无明显畸形的受试者100例,获取三维颜面数据,由专家进行颜面标志点(21个)和正中矢状平面的标注。以上述其中80例受试者三维颜面数据作为训练集数据,训练并建立本研究的MSH-CNN算法模型。以其余20例作为测试集数据,由训练后的深度学习算法自动确定每例数据的三维颜面解剖标志点(21个), 并评价算法标点与专家标点间“定点误差”。将MSH-CNN自动确定的三维颜面解剖标志点应用于本课题组前期研究建立的赋权普氏分析算法,可自动构建出20例受试者的三维颜面正中矢状平面。计算MSH-CNN结合赋权普氏分析算法构建的正中矢状平面与专家正中矢状平面间“角度误差”,评价三维颜面正中矢状平面自动构建方法的效果。结果: 针对20例面部无明显畸形的受试者,基于MSH-CNN和赋权普氏分析算法构建正中矢状平面与专家平面间的角度误差平均为0.73°±0.50°,其中MSH-CNN自动确定颜面21个解剖标志点的定点误差平均为(1.13±0.24) mm,眶区定点误差最大平均为(1.31±0.54) mm,鼻区定点误差最小平均为(0.79±0.36) mm。结论: 将深度学习算法与赋权普氏分析算法结合应用,实现了三维颜面正中矢状平面的全自动构建,初步达到了临床专家的构建效果,为自主知识产权的软件开发奠定了基础。     

基于歧义消解的文献主题标引系统研究与实现

目的：通过主题标引的歧义消解机制，有效过滤歧义概念，从而提高文献主题自动标引的准确性。方法：基于《STKOS超级词表》，构建《国家科技图书文献中心期刊分类-STKOS范畴对应表》，通过概念与文献的领域一致性原则过滤歧义概念，结合标注词典生成、术语原形化、通用概念过滤、概念遴选等过程优化外文科技文献主题标引系统。结果：在科技期刊数据集上主题标引评测的准确率为77.53%，召回率为73.25%，F值为75.33%。结论：通过歧义消解机制能够有效提高外文科技文献的主题标引效果。     

应用判别分析模型预测女性盆腔肿物性质的研究

目的探讨依据患者肿瘤标志物及一般资料建立的判别预测模型对于盆腔肿物性质诊断的价值。方法回顾性分析124例因盆腔肿物住院接受手术的患者,依据术后病理分为卵巢良性肿物组85例,子宫平滑肌瘤组25例,卵巢恶性肿瘤组14例,将所有的数据随机分成两个数据集,分别为训练集104例和验证集20例。收集上述所有患者的一般资料并检测其肿瘤标志物水平,建立判别预测模型并对其进行验证。结果本研究将训练集通过以下变量建立了判别预测模型-1: 年龄、孕次、产次、身高、体重、BMI、初潮年龄、绝经与否、CA153、CA125、AFP、CEA、CYFRA21-1、SCCAg、CA199、抑制素B,通过交叉验证得到其准确率为91.3%。同时,将上述数据采用Logistic回归分析,寻找有统计学意义的变量,发现年龄、孕次、抑制素B、CA125等统计资料差异有统计学意义(P<0.05)。再将上述变量建立判别预测模型-2,通过交叉验证后得到其准确率为93.3%。最后应用验证集将两个模型进行验证。结论判别预测模型-1和判别预测模型-2均可用于盆腔肿物性质的预测,但后者对盆腔肿物性质的预测具有数据简化、灵敏度高等优势,对于术前临床辅助诊断具有重要的临床意义。     

固定床反应器基于改进混合模型的串级推断控制

提出了非绝热式固定床反应器基于改进混合模型的串级非线性推断控制策略。道德提出了邻二甲苯氧化非绝热式固定床反应器基于改进混合模型的推断估计器;然后设计了基于新型非线性自适应控制算法的副控制器和基于程序变增益ＰＩＤ的主控制器,同时设计了新型非线性滤波器以解决该反应器主回路比副回路响应快的问题;最后建立了该固定床反应器基于改进混合模型的串级非线性推断控制系统。仿真结果表明该推断估计器具有良好的静态、动态     

Hybrid methods for improving information access in clinical documents: concept,assertion, and relation identification

Objective

This paper describes the approaches the authors developed while participating in the i2b2/VA 2010 challenge to automatically extract medical concepts and annotate assertions on concepts and relations between concepts.

Design

The authors''approaches rely on both rule-based and machine-learning methods. Natural language processing is used to extract features from the input texts; these features are then used in the authors'' machine-learning approaches. The authors used Conditional Random Fields for concept extraction, and Support Vector Machines for assertion and relation annotation. Depending on the task, the authors tested various combinations of rule-based and machine-learning methods.

Results

The authors''assertion annotation system obtained an F-measure of 0.931, ranking fifth out of 21 participants at the i2b2/VA 2010 challenge. The authors'' relation annotation system ranked third out of 16 participants with a 0.709 F-measure. The 0.773 F-measure the authors obtained on concept extraction did not make it to the top 10.

Conclusion

On the one hand, the authors confirm that the use of only machine-learning methods is highly dependent on the annotated training data, and thus obtained better results for well-represented classes. On the other hand, the use of only a rule-based method was not sufficient to deal with new types of data. Finally, the use of hybrid approaches combining machine-learning and rule-based approaches yielded higher scores.     

A Three-Stage Expert System Based on Support Vector Machines for Thyroid Disease Diagnosis

In this paper, we present a three-stage expert system based on a hybrid support vector machines (SVM) approach to diagnose thyroid disease. Focusing on feature selection, the first stage aims at constructing diverse feature subsets with different discriminative capability. Switching from feature selection to model construction, in the second stage, the obtained feature subsets are fed into the designed SVM classifier for training an optimal predictor model whose parameters are optimized by particle swarm optimization (PSO). Finally, the obtained optimal SVM model proceeds to perform the thyroid disease diagnosis tasks using the most discriminative feature subset and the optimal parameters. The effectiveness of the proposed expert system (FS-PSO-SVM) has been rigorously evaluated against the thyroid disease dataset, which is commonly used among researchers who use machine learning methods for thyroid disease diagnosis. The proposed system has been compared with two other related methods including the SVM based on the Grid search technique (Grid-SVM) and the SVM based on Grid search and principle component analysis (PCA-Grid-SVM) in terms of their classification accuracy. Experimental results demonstrate that FS-PSO-SVM significantly outperforms the other ones. In addition, Compared to the existing methods in previous studies, the proposed system has achieved the highest classification accuracy reported so far by 10-fold cross-validation (CV) method, with the mean accuracy of 97.49% and with the maximum accuracy of 98.59%. Promisingly, the proposed FS-PSO-SVM expert system might serve as a new candidate of powerful tools for diagnosing thyroid disease with excellent performance.     

音乐训练和美术训练对大学生字母-语音视听整合的影响

目的考察音乐训练和美术训练对大学生的字母-语音视听整合的影响。方法以视觉(字母)和听觉(语音)为刺激内容,对音乐生、美术生和普通大学生(每组n=30)进行辨别任务,采用SPSS 17.0对反应时和击中率进行方差分析和竞争模型(Race model)分析。结果(1)反应时:结果发现组别与刺激类型的交互作用显著(F=7.89,P<0.01)。简单效应分析发现,所有被试对一致的视听刺激(字母B读音B)反应更快,出现冗余信号效应。(2)Race model分析:音乐组视听整合的时间窗口为120~130 ms,美术组为130~190 ms,普通组为120~170 ms(P<0.01)。音乐组Race model曲线下正值面积(18.95)显著小于美术组(159.43)和普通组(125.01)(P<0.01),美术组和普通组之间差异无统计学意义(P=0.13)。结论音乐训练对字母-语音视听整合有调节作用,而美术训练对其没有显著影响。     

Using Autoencoders for Mammogram Compression

This paper presents the results obtained for medical image compression using autoencoder neural networks. Since mammograms (medical images) are usually of big sizes, training of autoencoders becomes extremely tedious and difficult if the whole image is used for training. We show in this paper that the autoencoders can be trained successfully by using image patches instead of the whole image. The compression performances of different types of autoencoders are compared based on two parameters, namely mean square error and structural similarity index. It is found from the experimental results that the autoencoder which does not use Restricted Boltzmann Machine pre-training yields better results than those which use this pre-training method.     

MIARS: A Medical Image Retrieval System

The next generation of medical information system will integrate multimedia data to assist physicians in clinical decision-making, diagnoses, teaching, and research. This paper describes MIARS (Medical Image Annotation and Retrieval System). MIARS not only provides automatic annotation, but also supports text based as well as image based retrieval strategies, which play important roles in medical training, research, and diagnostics. The system utilizes three trained classifiers, which are trained using training images. The goal of these classifiers is to provide multi-level automatic annotation. Another main purpose of the MIARS system is to study image semantic retrieval strategy by which images can be retrieved according to different levels of annotation.     

A Computer Aided Diagnosis System for Thyroid Disease Using Extreme Learning Machine

In this paper, we present an effective and efficient computer aided diagnosis (CAD) system based on principle component analysis (PCA) and extreme learning machine (ELM) to assist the task of thyroid disease diagnosis. The CAD system is comprised of three stages. Focusing on dimension reduction, the first stage applies PCA to construct the most discriminative new feature set. After then, the system switches to the second stage whose target is model construction. ELM classifier is explored to train an optimal predictive model whose parameters are optimized. As we known, the number of hidden neurons has an important role in the performance of ELM, so we propose an experimental method to hunt for the optimal value. Finally, the obtained optimal ELM model proceeds to perform the thyroid disease diagnosis tasks using the most discriminative new feature set and the optimal parameters. The effectiveness of the resultant CAD system (PCA-ELM) has been rigorously estimated on a thyroid disease dataset which is taken from UCI machine learning repository. We compare it with other related methods in terms of their classification accuracy. Experimental results demonstrate that PCA-ELM outperforms other ones reported so far by 10-fold cross-validation method, with the mean accuracy of 97.73% and with the maximum accuracy of 98.1%. Besides, PCA-ELM performs much faster than support vector machines (SVM) based CAD system. Consequently, the proposed method PCA-ELM can be considered as a new powerful tools for diagnosing thyroid disease with excellent performance and less time.     

Design of tele-educational system for interpreting dental tumor images

Despite the fact that oral cancer is usually diagnosed with the naked eye, dental radiology can play a significant role in the process of diagnosis, treatment planning, assessment of response to treatment, and prognosis. This paper will discuss a prototype tele-educational system to support dental radiology training programs, which will help dental students get the skills necessary for interpreting dental tumor images. The system consists of a central database, an online case annotation tool and a case demonstration tool. The annotation tool allows dental faculty to integrate image findings with related clinical information, and to prepare high-quality teaching cases. The tele-educational system can improve and reinforce dental students’ and dentists’ skills in analyzing dental tumor images. The design of tele-educational system could serve as a model for self-evaluation of interpreting skills in dental tumor images as part of Continuous Medical Education (CME) in the future.     

基于知识图谱和模型融合的医疗问答系统的构建

基于语义解析和规则匹配融合的模型，利用少量的语义训练语料，以中文医疗知识图谱为知识基础，构建中文医疗问答系统，解决医疗领域中文语料缺乏且标注难度大的问题。该系统由语义解析模块（SPM）和答案查询模块（AQM）组成。其中，SPM由意图识别和命名实体识别组成，它们分别以BERT-TextCNN和BiLSTM-CRF模型为基础，融合了规则校正，其准确率较非融合模型分别提升9%和11%。实验结果表明，该系统回答准确率达到82%，具有较强的问题解答能力和一定的实用价值。     

84例天疱疮患者临床疗效分析和预后观察

[[摘要] 目的 总结天疱疮患者的治疗方案,分析其治疗效果及预后。方法 回顾分析84例天疱疮患者的临床资料,治疗药物均以糖皮质激素（以下简称激素）为主,45例患者同时加用免疫抑制剂（其中3例因治疗效果差进行了自体外周血干细胞移植）,观察激素用量、治疗效果、副作用以及预后情况。结果 重症组患者激素初始量高于中症组患者;激素联合免疫抑制剂组治疗效果优于激素组;随访患者中9例因激素减量过快复发;3例进行了干细胞移植的患者均获得满意疗效,已停用激素。结论 天疱疮病情的严重程度是决定激素用量的关键。激素联合免疫抑制剂治疗天疱疮疗效更佳。对激素和免疫抑制剂治疗效果不理想者可选择自体外周血造血干细胞移植,后者可能是治愈天疱疮的一种有效方法,有待进一步探索。