融合深度网络和浅层纹理特征的甲状腺结节癌变超声图像诊断

目的 在甲状腺结节图像中对甲状腺结节进行良恶性分析,对于甲状腺癌的早期诊断有着重要的意义。随着医疗影像学的发展,大部分的早期甲状腺结节可以在超声图像中准确地检测出来,但对于结节的性质仍然缺乏准确的判断。因此,为实现更为准确的早期甲状腺结节良恶性超声图像诊断,避免不必要的针刺或其他病理活检手术、减轻病患生理痛苦和心理压力及其医疗费用,提出一种基于深度网络和浅层纹理特征融合的甲状腺结节良恶性分类新算法。方法 本文提出的甲状腺结节分类算法由4步组成。首先对超声图像进行尺度配准、人工标记以及图像复原去除以增强图像质量。然后,对增强的图像进行数据扩展,并作为训练集对预训练过的GoogLeNet卷积神经网络进行迁移学习以提取图像中的深度特征。同时,提取图像的旋转不变性局部二值模式（LBP）特征作为图像的纹理特征。最后,将深度特征与图像的纹理特征相融合并输入至代价敏感随机森林分类器中对图像进行良恶性分类。结果 本文方法在标准的甲状腺结节癌变数据集上对甲状腺结节图像取得了正确率99.15%,敏感性99.73%,特异性95.85%以及ROC曲线下面积0.997 0的的好成绩,优于现有的甲状腺结节图像分类方法。结论 实验结果表明,图像的深度特征可以描述医疗超声图像中病灶的整体感官特征,而浅层次纹理特征则可以描述超声图像的边缘、灰度分布等特征,将二者统一的融合特征则可以更为全面地描述图像中病灶区域与非病灶区域之间的差异以及不同病灶性质之间的差异。因此,本文方法可以准确地对甲状腺结节进行分类从而避免不必要手术、减轻病患痛苦和压力。     

基于不同超声成像的甲状腺结节良恶性判别

为实现更为准确的甲状腺结节良恶性超声图像诊断,避免不必要的穿刺或活检手术,提出了一种基于卷积神经网络（CNN）的常规超声成像和超声弹性成像的特征结合方法,提高了甲状腺结节良恶性分类准确率。首先,卷积网络模型在大规模自然图像数据集上完成预训练,并通过迁移学习的方式将特征参数迁移到超声图像域用以生成深度特征并处理小样本。然后,结合常规超声成像和超声弹性成像的深度特征图形成混合特征空间。最后,在混合特征空间上完成分类任务,实现了一个端到端的卷积网络模型。在1156幅图像上进行实验,所提方法的准确率为0.924,高于其他单一数据源的方法。实验结果表明,浅层卷积共享图像的边缘纹理特征,高层卷积的抽象特征与具体的分类任务相关,使用迁移学习的方法可以解决数据样本不足的问题;同时,弹性超声影像可以对甲状腺结节的病灶硬度进行客观的量化,结合常规超声的纹理轮廓特征,二者融合的混合特征可以更全面地描述不同病灶之间的差异。所提方法可以高效准确地对甲状腺结节进行良恶性分类,减轻患者痛苦,给医生提供更为准确的辅助诊断信息。     

Analysis of spatial variability using geostatistical functions for diagnosis of lung nodule in computerized tomography images

This paper analyzes four geostatistical functions —semivariogram, semimadogram, covariogram, and correlogram—with the purpose of characterizing lung nodules as malignant or benign in computerized tomography images. The tests described in this paper were carried out using a sample of 30 nodules, 24 benign and 6 malignant. Stepwise discriminant analysis was used to determine which combination of measures were best able to discriminate between the benign and malignant nodules. Then, a linear discriminant analysis procedure was performed using the selected features to evaluate the ability of these features to predict the classification for each nodule. A leave-one-out procedure was used to provide a less biased estimate of the linear discriminator’s performance. All analyzed functions have value area under receiver operation characteristic (ROC) curve above 0.800, which means results with accuracy between good and excellent. The preliminary results of this approach are very promising in characterizing nodules using geostatistical functions.     

几种机器学习方法在黑色素瘤计算机辅助诊断中的性能比较

黑色素瘤的计算机辅助诊断是基于激光共聚焦扫描显微镜（CLSM）皮肤图像纹理特征, 并引入机器学习的技术, 为临床应用研发的一种能够准确、有效地识别在体恶性黑色素瘤新医学诊断方法, 将常用的基于机器学习的ID3、分类与回归树（CART）和AdaBoost三种算法应用于良恶性黑色素瘤图像的特征识别, 并对各种学习方法的性能进行比较。实验结果表明, AdaBoost算法具有较好的分类识别性能, 不但提高了恶性黑色素瘤早期诊断的准确度, 降低了良性黑色素瘤的误诊率, 而且为临床上早期发现和诊断提供了客观依据。     

基于改进ReliefF与k-means算法的良恶性肺结节分类模型

肺结节是肺癌的症状.在CT图像中,肺结节的形状和大小常被用来进行肺癌的诊断,然而良性和恶性结节的鉴别对于疾病的治疗具有重要意义.由于良恶性结节的边缘纹理特征区别大,因此本文首先利用基于改进的边缘检测算子的灰度-梯度共生矩阵(GGCM)提取小梯度优势、灰度分布不均匀性、能量、灰度熵、梯度熵、混合熵、逆差距、相关性等肺部CT图像的14种纹理特征.然后利用改进的ReliefF算法去除作用小的特征,保留重要特征的特征权重值.最后将重要特征的权重值应用于改进距离度量准则的k-means算法中进行良恶性结节的分类.应用本文算法在LIDC数据集上实验,实验分析结果表明,14种纹理特征对于结节良恶性的分类能力并不相同,而灰度差、梯度差、能量、小梯度优势、相关性、灰度熵、混合熵、逆差矩的组合得到的良恶性肺结节分类效果最好,最终实现了良性结节83.46%,恶性结节95.02%的识别率,可在临床应用中辅助医生进行肺结节的良恶性诊断.     

肺结节球表面网格向量化特征分类

目的 基于球谐函数与容斥映射算法向量化球面表面纹理与结节形状用以进行胸部CT图像肺结节良恶性判定。区别于基于深度学习解决肺结节良恶性筛查的方法,目前方法多集中于框架改进而忽略了数据预处理,文中所提方法旨在对球面纹理与结节形状进行向量表达,使其可以输入深度森林进行特征分类训练。方法 首先采用辽宁中医药大学附属医院数据,通过3维重构获得3维肺结节图像。其次使用球谐函数与容斥映射算法在保留空间信息的同时将纹理以网格方式映射到标准球面上。再次使用网格-LBP与映射形变能量分别完成对球面纹理与结节形状信息的构建。最后提出一种基于网格的多粒度扫描方法对深度森林训练框架进行改进,并将向量化后的纹理和形状特征加入到改进的深度森林训练框架中进行实验验证。结果 通过大量的实验结果验证,在准确率（ACC）、特异度（SPE）、敏感度（SEN）和受试者工作特征曲线下的面积（AUC）4个衡量指标下,本文方法具有优于现存先进方法的表现,其中ACC、SPE、SEN和AUC分别达到76.06%、69.46%、88.46%和0.84。结论 基于球谐函数与容斥映射算法可成功地对肺结节表面和形状两个特征进行向量化并训练,不仅考虑了数据预处理,而且通过两个特征对肺结节良恶性检测的准确率要高于传统1个特征检测的结果,同时也为3维模型中特征的提取及向量化提供了一个有效的方法。     

Automatic classification for solitary pulmonary nodule in CT image by fractal analysis based on fractional Brownian motion model

Perfusion computed tomography (CT) method has been used to differentiate malignant pulmonary nodules from benign nodules based on the assessment for the change of the CT attenuation value within the pulmonary nodules. Instead of using the change of the CT attenuation value, a set of fractal features based on fractional Brownian motion model is proposed in this paper to automatically distinguish malignant nodules from benign nodules. In a set of 107 CT images from 107 different patients with each image containing a solitary pulmonary nodule, our experimental results obtained from a support vector machine classifier show that the accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and the area under the ROC curve are 83.11%, 90.92%, 71.70%, 80.05%, 87.52%, and 0.8437, respectively, by using the proposed fractal-based feature set. Such a result outperforms the conventional method of using the change of the CT attenuation value as the feature for classification. When combining this conventional method with our proposed fractal-based method, the accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and the area under the ROC curve can be promoted to 88.82%, 93.92%, 82.90%, 87.30%, 90.20%, and 0.9019, respectively. In other words, a high performance of pulmonary nodule classification can be achieved with a single post-contrast CT scan.     

基于遗传算法和BP神经网络的孤立性肺结节分类算法

为了提高计算机辅助诊断系统中孤立性肺结节的良恶性诊断的准确性,提出了一种基于遗传算法和BP神经网的分类算法。该算法针对BP神经网络容易陷入局部最优的问题,综合考虑孤立性肺结节的医学诊断特性,采用遗传算法对基于BP神经网络的分类器进行优化,并通过对PET/CT图像进行处理,提取病灶的功能特征、结构特征以及临床信息作为神经网络分类器的输入样本,实现孤立性肺结节的良恶性分类。对医院以及网络公共数据库中的大量实验数据进行分类实验,结果表明优化后的算法在分类准确性上有较大的提高,说明该方法在肺结节临床分类方面是有效的。     

基于放射组学特征的胃肠道间质瘤的分类预测

胃肠道间质瘤(GastroIntestinal Stromal Tumors,GIST)是常见的胃肠道肿瘤,具有非定向分化特征,缺乏特异性,且具有恶性潜能,所以GIST的良恶性诊断是临床较为关注的问题。然而,病理活检及CT检查等临床鉴别手段在研究肿瘤异质性方面存在一定困难。文中提出一种基于CT图像提取大量量化的放射组学特征并利用SVM分类器对GIST良恶性进行分类预测的非侵入式方法。首先,应用放射组学方法对120个患有GIST的病人的CT图像肿瘤区域分别提取4个非纹理特征和43个纹理特征。 然后,应用基于ReliefF的前向选择算法进行特征选择,再用最佳特征子集训练得到的SVM分类器来对GIST良恶性进行分类预测。实验中,共有14个纹理特征入选最佳特征子集,且SVM分类模型对GIST良恶性分类的AUC、准确率、敏感性、特异性在训练集中分别为0.9949,0.9277,0.9537,0.9018;在测试集中分别为0.8524,0.8313,0.8197,0.8420。该方法以放射组学的研究方法建立的模型,为GIST良恶性预测提供了一种非入侵式的检测手段,有望成为一种辅助诊断工具,以提高临床GIST良恶性诊断的准确率。     

Diagnosis of lung nodule using Moran’s index and Geary’s coefficient in computerized tomography images

This paper analyzes the application of Moran’s index and Geary’s coefficient to the characterization of lung nodules as malignant or benign in computerized tomography images. The characterization method is based on a process that verifies which combination of measures, from the proposed measures, has been best able to discriminate between the benign and malignant nodules using stepwise discriminant analysis. Then, a linear discriminant analysis procedure was performed using the selected features to evaluate the ability of these in predicting the classification for each nodule. In order to verify this application we also describe tests that were carried out using a sample of 36 nodules: 29 benign and 7 malignant. A leave-one-out procedure was used to provide a less biased estimate of the linear discriminator’s performance. The two analyzed functions and its combinations have provided above 90% of accuracy and a value area under receiver operation characteristic (ROC) curve above 0.85, that indicates a promising potential to be used as nodules signature measures. The preliminary results of this approach are very encouraging in characterizing nodules using the two functions presented.

Comparison of multiwavelet, wavelet, Haralick, and shape features for microcalcification classification in mammograms

We present an evaluation and comparison of the performance of four different texture and shape feature extraction methods for classification of benign and malignant microcalcifications in mammograms. For 103 regions containing microcalcification clusters, texture and shape features were extracted using four approaches: conventional shape quantifiers; co-occurrence-based method of Haralick; wavelet transformations; and multi-wavelet transformations. For each set of features, most discriminating features and their optimal weights were found using real-valued and binary genetic algorithms (GA) utilizing a k-nearest-neighbor classifier and a malignancy criterion for generating ROC curves for measuring the performance. The best set of features generated areas under the ROC curve ranging from 0.84 to 0.89 when using real-valued GA and from 0.83 to 0.88 when using binary GA. The multi-wavelet method outperformed the other three methods, and the conventional shape features were superior to the wavelet and Haralick features.     

基于混合受限玻尔兹曼机的肺结节良恶性诊断

针对传统计算机辅助诊断中肺结节的特征提取方法依靠人工设计、操作复杂、识别率低等问题,提出了一种基于混合受限玻尔兹曼机的肺结节良恶性诊断方法。首先采用多层无监督卷积受限玻尔兹曼机自动对肺结节图像进行特征学习,然后利用分类受限玻尔兹曼机对获得的特征进行良恶性分类。为避免分类受限玻尔兹曼机在训练中出现的特征同质化问题,引入了交叉熵稀疏惩罚对其进行优化。实验结果表明,该方法有效避免了手动特征提取的复杂性,在肺结节良恶性分类的准确率、敏感性、特异性、ROC曲线下面积值上均优于传统诊断方法。     

基于多尺度融合的甲状腺结节图像特征提取

甲状腺结节是一种常见的多发病,超声技术是该疾病首选的检查方法。在超声图像中提取区分甲状腺结节良恶性的纹理特征并进行判别具有广阔的临床应用前景。双树复小波变换（Dual-tree complex wavelet transform,DT-CWT）和Gabor小波是纹理特征提取的常用方法。本文提出一种基于多尺度的DT-CWT和Gabor特征融合的甲状腺结节识别方法。该方法首先通过高斯金字塔将甲状腺超声图像分解到多尺度空间,然后提取图像的DT-CWT和Gabor的多尺度特征,最后实现特征融合。通过应用支持向量机（Support vector machine,SVM）分类器实现分类,验证特征提取方法的有效性。实验结果表明,本文提出的方法能达到较高的识别率。     

Diagnosis of lung nodule using semivariogram and geometric measures in computerized tomography images

This paper uses the geostatistical function - semivariogram and a set of 3D geometric measures - sphericity index, convexity index, extrinsic and intrinsic curvature index and surface type, to characterize lung nodules as malignant or benign in computerized tomography images. Based on a sample of 31 nodules, 25 benign and 6 malignant, these methods are first analyzed individually and then jointly, with techniques for classification and analysis (stepwise discriminant analysis, leave-one-out and ROC curve). We have concluded that the individual measures and their combinations produce good results in the diagnosis of lung nodules.     

Toward a combined tool to assist dermatologists in melanoma detection from dermoscopic images of pigmented skin lesions

In this paper we propose a machine learning approach to classify melanocytic lesions as malignant or benign, using dermoscopic images. The lesion features used in the classification framework are inspired on border, texture, color and structures used in popular dermoscopy algorithms performed by clinicians by visual inspection. The main weakness of dermoscopy algorithms is the selection of a set of weights and thresholds, that appear not to be robust or independent of population. The use of machine learning techniques allows to overcome this issue. The proposed method is designed and tested on an image database composed of 655 images of melanocytic lesions: 544 benign lesions and 111 malignant melanoma. After an image pre-processing stage that includes hair removal filtering, each image is automatically segmented using well known image segmentation algorithms. Then, each lesion is characterized by a feature vector that contains shape, color and texture information, as well as local and global parameters. The detection of particular dermoscopic patterns associated with melanoma is also addressed, and its inclusion in the classification framework is discussed. The learning and classification stage is performed using AdaBoost with C4.5 decision trees. For the automatically segmented database, classification delivered a specificity of 77% for a sensitivity of 90%. The same classification procedure applied to images manually segmented by an experienced dermatologist yielded a specificity of 85% for a sensitivity of 90%.     

Information fusion approach to microcalcification characterization

The paper presents an information fusion-based approach to one of the most challenging problems in mammogram interpretation: the problem of characterizing mammographic microcalcifications as benign or malignant.There are two categories of methods typically used for designing decision aids for diagnosis of microcalcifications: computer vision methods employing intensity-based features automatically extracted from images and methods using mammogram characteristics considered by human experts. The achieved recognition accuracy of both types of methods is not yet sufficient for them to be utilized in clinical practice. The paper introduces a hybrid system combining decisions of classifiers utilizing both domain knowledge-based and intensity-based features within the framework of the Evidence theory. The system comprises a hierarchical evidential classifier employing a combination of texture features of individual microcalcifications and a neural network employing cluster features observed and described by a radiologist. The results of a pilot study have shown that a false alarm rate of the hybrid system is lower than the false alarm rate of each single classifier used in the combination as well as that of the radiologists participated in the study.     

Application of support-vector-machine-based method for feature selection and classification of thyroid nodules in ultrasound images

Most thyroid nodules are heterogeneous with various internal components, which confuse many radiologists and physicians with their various echo patterns in ultrasound images. Numerous textural feature extraction methods are used to characterize these patterns to reduce the misdiagnosis rate. Thyroid nodules can be classified using the corresponding textural features. In this paper, six support vector machines (SVMs) are adopted to select significant textural features and to classify the nodular lesions of a thyroid. Experiment results show that the proposed method can correctly and efficiently classify thyroid nodules. A comparison with existing methods shows that the feature-selection capability of the proposed method is similar to that of the sequential-floating-forward-selection (SFFS) method, while the execution time is about 3-37 times faster. In addition, the proposed criterion function achieves higher accuracy than those of the F-score, T-test, entropy, and Bhattacharyya distance methods.     

多模态融合下长时程肺部病灶良恶性预测方法

为了更精确、全面地表征各时期肺部医学影像中病灶特征的变化与发展规律，研究在时间纵向维度上预测肺结节的演变方式，构建了一种多模态特征融合下不同时期肺部病灶良恶性预测模型。根据病人不同时期的序列CT图像，提取肺部病灶的传统特征与深度特征，构造多模态特征；通过神经网络对多模态特征进行相关性快速融合；利用长短时记忆方法学习不同时期具有时间特征的肺部病灶特征向量，构建一个双向长短时记忆网络对病灶进行良恶性预测。实验表明，所提方法准确率为92.8%，比传统方法有所提高，可以实现有效预测。     

基于改进LVQ算法的肺结节识别与分类

目前,肺癌的是发病率最高的肿瘤,若能在早期发现癌变并进行相应治疗,将极大的提高患者的生存率。肺癌的症状在早期表现为肺结节。以提高肺结节检测识别率并进行良恶性分类为目的,提出了一种改进的LVQ分类器算法。首先使用C-V算法对原始图像进行肺实质分割,再使用最优阈值法进行感兴趣区域提取,并进行特征提取和特征归一化。使用多次聚类算法检测肺结节。使用基于改进的LVQ分类器进行肺结节的良恶性进行分类。利用改进后的LVQ分类器在LIDC数据集上进行实验,得到了对良性结节的确诊率为87.3%,对恶性结节的确诊率为80.8%。实验结果表明,改进后的算法在良恶性结节分类上具有较高的确诊率,有助于提高医生的工作效率,实现肺结节的辅助发现。     

一种基于直方图特征和AdaBoost的图像中的文字定位算法

图像中的文字自动定位是计算机视觉领域中的一个新兴研究热点。为了使得定位算法能够适应不同类型的图像和文字，根据文字所具有的特殊纹理属性，提出了一种具有普适能力的基于直方图特征和AdaBoost的文字定位算法。该算法首先通过提取对文字具有较强鉴别能力的直方图特征和引入AdaBoost算法来设计级联结构的纹理分类器；然后用该分类器的概率输出来生成文字概率图；在此基础上再通过CAMSHIFT算法得到最终的定位结果。实验结果表明，该算法具有较强的鲁棒性，能够适应文字在语种、字体、尺度等方面的变化，在不同类型的图像中都能得到较好的定位结果。