Computer-aided breast cancer detection and diagnosis of massesusing difference of Gaussians and derivative-based feature saliency

A new model-based vision (MBV) algorithm is developed to find regions of interest (ROI's) corresponding to masses in digitized mammograms and to classify the masses as malignant/benign. The MBV algorithm is comprised of 5 modules to structurally identify suspicious ROI's, eliminate false positives, and classify the remaining as malignant or benign. The focus of attention module uses a difference of Gaussians (DoG) filter to highlight suspicious regions in the mammogram. The index module uses tests to reduce the number of nonmalignant regions from 8.39 to 2.36 per full breast image. Size, shape, contrast, and Laws texture features are used to develop the prediction module's mass models. Derivative-based feature saliency techniques are used to determine the best features for classification. Nine features are chosen to define the malignant/benign models. The feature extraction module obtains these features from all suspicious ROI's. The matching module classifies the regions using a multilayer perceptron neural network architecture to obtain an overall classification accuracy of 100% for the segmented malignant masses with a false-positive rate of 1.8 per full breast image. This system has a sensitivity of 92% for locating malignant ROI's. The database contains 272 images (12 b, 100 μm) with 36 malignant and 53 benign mass images. The results demonstrate that the MBV approach provides a structured order of integrating complex stages into a system for radiologists     

Classification of mass and normal breast tissue: a convolution neural network classifier with spatial domain and texture images

The authors investigated the classification of regions of interest (ROI's) on mammograms as either mass or normal tissue using a convolution neural network (CNN). A CNN is a backpropagation neural network with two-dimensional (2-D) weight kernels that operate on images. A generalized, fast and stable implementation of the CNN was developed. The input images to the CNN were obtained from the ROI's using two techniques. The first technique employed averaging and subsampling. The second technique employed texture feature extraction methods applied to small subregions inside the ROI. Features computed over different subregions were arranged as texture images, which were subsequently used as CNN inputs. The effects of CNN architecture and texture feature parameters on classification accuracy were studied. Receiver operating characteristic (ROC) methodology was used to evaluate the classification accuracy. A data set consisting of 168 ROIs containing biopsy-proven masses and 504 ROI's containing normal breast tissue was extracted from 168 mammograms by radiologists experienced in mammography. This data set was used for training and testing the CNN. With the best combination of CNN architecture and texture feature parameters, the area under the test ROC curve reached 0.87, which corresponded to a true-positive fraction of 90% at a false positive fraction of 31%. The authors' results demonstrate the feasibility of using a CNN for classification of masses and normal tissue on mammograms.     

Detection of daily activities and sports with wearable sensors in controlled and uncontrolled conditions

Physical activity has a positive impact on people's well-being, and it may also decrease the occurrence of chronic diseases. Activity recognition with wearable sensors can provide feedback to the user about his/her lifestyle regarding physical activity and sports, and thus, promote a more active lifestyle. So far, activity recognition has mostly been studied in supervised laboratory settings. The aim of this study was to examine how well the daily activities and sports performed by the subjects in unsupervised settings can be recognized compared to supervised settings. The activities were recognized by using a hybrid classifier combining a tree structure containing a priori knowledge and artificial neural networks, and also by using three reference classifiers. Activity data were collected for 68 h from 12 subjects, out of which the activity was supervised for 21 h and unsupervised for 47 h. Activities were recognized based on signal features from 3-D accelerometers on hip and wrist and GPS information. The activities included lying down, sitting and standing, walking, running, cycling with an exercise bike, rowing with a rowing machine, playing football, Nordic walking, and cycling with a regular bike. The total accuracy of the activity recognition using both supervised and unsupervised data was 89% that was only 1% unit lower than the accuracy of activity recognition using only supervised data. However, the accuracy decreased by 17% unit when only supervised data were used for training and only unsupervised data for validation, which emphasizes the need for out-of-laboratory data in the development of activity-recognition systems. The results support a vision of recognizing a wider spectrum, and more complex activities in real life settings.     

Application of neural networks to radar image classification

A number of methods have been developed to classify ground terrain types from fully polarimetric synthetic aperture radar (SAR) images, and these techniques are often grouped into supervised and unsupervised approaches. Supervised methods have yielded higher accuracy than unsupervised techniques, but suffer from the need for human interaction to determine classes and training regions. In contrast, unsupervised methods determine classes automatically, but generally show limited ability to accurately divide terrain into natural classes. In this paper, a new terrain classification technique is introduced to determine terrain classes in polarimetric SAR images, utilizing unsupervised neural networks to provide automatic classification, and employing an iterative algorithm to improve the performance. Several types of unsupervised neural networks are first applied to the classification of SAR images, and the results are compared to those of more conventional unsupervised methods. Results show that one neural network method-Learning Vector Quantization (LVQ)-outperforms the conventional unsupervised classifiers, but is still inferior to supervised methods. To overcome this poor accuracy, an iterative algorithm is proposed where the SAR image is reclassified using a maximum likelihood (ML) classifier. It is shown that this algorithm converges, and significantly improves classification accuracy     

Statistical textural features for detection of microcalcificationsin digitized mammograms

Clustered microcalcifications on X-ray mammograms are an important sign for early detection of breast cancer. Texture-analysis methods can be applied to detect clustered microcalcifications in digitized mammograms. In this paper, a comparative study of texture-analysis methods is performed for the surrounding region-dependence method, which has been proposed by the authors, and conventional texture-analysis methods, such as the spatial gray-level dependence method, the gray-level run-length method, and the gray-level difference method. Textural features extracted by these methods are exploited to classify regions of interest (ROI's) into positive ROI's containing clustered microcalcifications and negative ROI's containing normal tissues. A three-layer backpropagation neural network is used as a classifier. The results of the neural network for the texture-analysis methods are evaluated by using a receiver operating-characteristics (ROC) analysis. The surrounding region-dependence method is shown to be superior to the conventional texture-analysis methods with respect to classification accuracy and computational complexity.     

Toward unsupervised adaptation of LDA for brain-computer interfaces

There is a step of significant difficulty experienced by brain-computer interface (BCI) users when going from the calibration recording to the feedback application. This effect has been previously studied and a supervised adaptation solution has been proposed. In this paper, we suggest a simple unsupervised adaptation method of the linear discriminant analysis (LDA) classifier that effectively solves this problem by counteracting the harmful effect of nonclass-related nonstationarities in electroencephalography (EEG) during BCI sessions performed with motor imagery tasks. For this, we first introduce three types of adaptation procedures and investigate them in an offline study with 19 datasets. Then, we select one of the proposed methods and analyze it further. The chosen classifier is offline tested in data from 80 healthy users and four high spinal cord injury patients. Finally, for the first time in BCI literature, we apply this unsupervised classifier in online experiments. Additionally, we show that its performance is significantly better than the state-of-the-art supervised approach.     

Supervised and unsupervised learning in radial basis functionclassifiers

The paper considers a number of strategies for training radial basis function (RBF) classifiers. A benchmark problem is constructed using ten-dimensional input patterns which have to be classified into one of three classes. The RBF networks are trained using a two-phase approach (unsupervised clustering for the first layer followed by supervised learning for the second layer), error backpropagation (supervised learning for both layers) and a hybrid approach. It is shown that RBF classifiers trained with error backpropagation give results almost identical to those obtained with a multilayer perceptron. Although networks trained with the two-phase approach give slightly worse classification results, it is argued that the hidden-layer representation of such networks is much more powerful, especially if it is encoded in the form of a Gaussian mixture model. During training, the number of subclusters present within the training database can be estimated: during testing, the activities in the hidden layer of the classification network can be used to assess the novelty of input patterns and thereby help to validate network outputs     

Application of neural networks for sea ice classification inpolarimetric SAR images

Several automatic methods have been developed to classify sea ice types from fully polarimetric synthetic aperture radar (SAR) images, and these techniques are generally grouped into supervised and unsupervised approaches. In previous work, supervised methods have been shown to yield higher accuracy than unsupervised techniques, but suffer from the need for human interaction to determine classes and training regions. In contrast, unsupervised methods determine classes automatically, but generally show limited ability to accurately divide terrain into natural classes. In this paper, a new classification technique is applied to determine sea ice types in polarimetric and multifrequency SAR images, utilizing an unsupervised neural network to provide automatic classification, and employing an iterative algorithm to improve the performance. The learning vector quantization (LVQ) is first applied to the unsupervised classification of SAR images, and the results are compared with those of a conventional technique, the migrating means method. Results show that LVQ outperforms the migrating means method, but performance is still poor. An iterative algorithm is then applied where the SAR image is reclassified using the maximum likelihood (ML) classifier. It is shown that this algorithm converges, and significantly improves classification accuracy. The new algorithm successfully identifies first-year and multiyear sea ice regions in the images at three frequencies. The results show that L- and P-band images have similar characteristics, while the C-band image is substantially different. Classification based on single features is also carried out using LVQ and the iterative ML method. It is found that the fully polarimetric classification provides a higher accuracy than those based on a single feature. The significance of multilook classification is demonstrated by comparing the results obtained using four-look and single-look classifications     

基于最佳样本标记的主动支持向量机学习策略

支持向量机通过随机选择标记的训练样本进行有监督学习,随着信息容量的增加和数据收集能力的提高,这需要耗费大量的标记工作量,给实际应用带来不少困难.本文提出了基于最佳样本标记的主动支持向量机学习策略:首先利用无监督聚类选择一个小规模的样本集进行标记,然后训练该标记样本集得到一个初始SVM分类器,然后利用该分类器主动选择最感兴趣的无标记样本进行标记,逐渐增加标记样本的数量,并在此基础上更新分类器,反复进行直到得到最佳性能的分类器.实验结果表明在基本不影响分类精度的情况下,主动学习选择的标记样本数量大大低于随机选择的标记样本数量,这大大降低了标记的工作量,而且训练速度同样有所提高.     

Detection of breast masses in mammograms by density slicing and texture flow-field analysis.

We propose a method for the detection of masses in mammographic images that employs Gaussian smoothing and sub-sampling operations as preprocessing steps. The mass portions are segmented by establishing intensity links from the central portions of masses into the surrounding areas. We introduce methods for analyzing oriented flow-like textural information in mammograms. Features based on flow orientation in adaptive ribbons of pixels across the margins of masses are proposed to classify the regions detected as true mass regions or false-positives (FPs). The methods yielded a mass versus normal tissue classification accuracy represented as an area (Az) of 0.87 under the receiver operating characteristics (ROCs) curve with a dataset of 56 images including 30 benign disease, 13 malignant disease, and 13 normal cases selected from the mini Mammographic Image Analysis Society database. A sensitivity of 81% was achieved at 2.2 FPs/image. Malignant tumor versus normal tissue classification resulted in a higher Az value of 0.9 under the ROC curve using only the 13 malignant and 13 normal cases with a sensitivity of 85% at 2.45 FPs/image. The mass detection algorithm could detect all the 13 malignant tumors successfully, but achieved a success rate of only 63% (19/30) in detecting the benign masses. The mass regions that were successfully segmented were further classified as benign or malignant disease by computing five texture features based on gray-level co-occurrence matrices (GCMs) and using the features in a logistic regression method. The features were computed using adaptive ribbons of pixels across the boundaries of the masses. Benign versus malignant classification using the GCM-based texture features resulted in Az = 0.79 with 19 benign and 13 malignant cases.     

Recognition of temporally changing action potentials in multiunitneural recordings

We present a method to iteratively train an artificial neural network (ANN) or other supervised pattern classifier in order to adaptively recognize and track temporally changing patterns. This method uses recently acquired data and the existing classifier to create new training sets, from which a new classifier is then trained. The procedure is repeated periodically using the most recently trained classifier. This scheme was evaluated by applying it to simulated situations that arise in chronic recordings of multiunit neural activity from peripheral nerves. The method was able to track the changes in these simulated chronic recordings and to provide better unit recognition rates than an unsupervised clustering method suited to this problem     

A New Neuro-Fuzzy Classifier with Application to On-Line Face Detection and Recognition

In this paper, we propose a neuro-fuzzy classifier (NEFCAR) that utilizes positive and negative rules with different rule importances to create the decision boundaries between different classes. The locally unsupervised and globally supervised training technique is adopted. The decision-based and approximation-based strategies are combined to provide a suitable amount of training for each training pattern. The reinforced and anti-reinforced learning rules are given with different weighting so that the training can be efficient and can reach convergence quickly. Moreover, NEFCAR can easily provide the confidence measure of each classification decision. Therefore, the rejection algorithm can be implemented in a straightforward manner. Noise tolerant training is conducted to improve the generalization performance and the confidence measure is adopted to avoid overtraining. The proposed classifier is applied to two applications. The first one is the Fisher iris data classification, and the second one is an on-line face detection and recognition application. Good classification results are obtained in both applications. In the on-line face detection and recognition system, two NEFCAR's are utilized: a two-class and a multi-class NEFCAR's are adopted to detect the face and recognize the face, respectively. The color of skin and the motion information are taken into consideration heuristically to improve the effectiveness of the face location algorithm.     

Unsupervised Quick Reduct Algorithm Using Rough Set Theory

Feature selection (FS) is a process to select features which are more informative.It is one of the important steps in knowledge discovery.The problem is that not all features are important.Some of the features may be redundant,and others may be irrelevant and noisy.The conventional supervised FS methods evaluate various feature subsets using an evaluation function or metric to select only those features which are related to the decision classes of the data under consideration.However,for many data mining applications,decision class labels are often unknown or incomplete,thus indicating the significance of unsupervised feature selection.However,in unsupervised learning,decision class labels are not provided.In this paper,we propose a new unsupervised quick reduct (QR) algorithm using rough set theory.The quality of the reduced data is measured by the classification performance and it is evaluated using WEKA classifier tool.The method is compared with existing supervised methods and the result demonstrates the efficiency of the proposed algorithm.     

Computer-aided diagnostic scheme for distinction between benign and malignant nodules in thoracic low-dose CT by use of massive training artificial neural network

Low-dose helical computed tomography (LDCT) is being applied as a modality for lung cancer screening. It may be difficult, however, for radiologists to distinguish malignant from benign nodules in LDCT. Our purpose in this study was to develop a computer-aided diagnostic (CAD) scheme for distinction between benign and malignant nodules in LDCT scans by use of a massive training artificial neural network (MTANN). The MTANN is a trainable, highly nonlinear filter based on an artificial neural network. To distinguish malignant nodules from six different types of benign nodules, we developed multiple MTANNs (multi-MTANN) consisting of six expert MTANNs that are arranged in parallel. Each of the MTANNs was trained by use of input CT images and teaching images containing the estimate of the distribution for the "likelihood of being a malignant nodule," i.e., the teaching image for a malignant nodule contains a two-dimensional Gaussian distribution and that for a benign nodule contains zero. Each MTANN was trained independently with ten typical malignant nodules and ten benign nodules from each of the six types. The outputs of the six MTANNs were combined by use of an integration ANN such that the six types of benign nodules could be distinguished from malignant nodules. After training of the integration ANN, our scheme provided a value related to the "likelihood of malignancy" of a nodule, i.e., a higher value indicates a malignant nodule, and a lower value indicates a benign nodule. Our database consisted of 76 primary lung cancers in 73 patients and 413 benign nodules in 342 patients, which were obtained from a lung cancer screening program on 7847 screenees with LDCT for three years in Nagano, Japan. The performance of our scheme for distinction between benign and malignant nodules was evaluated by use of receiver operating characteristic (ROC) analysis. Our scheme achieved an Az (area under the ROC curve) value of 0.882 in a round-robin test. Our scheme correctly identified 100% (76/76) of malignant nodules as malignant, whereas 48% (200/413) of benign nodules were identified correctly as benign. Therefore, our scheme may be useful in assisting radiologists in the diagnosis of lung nodules in LDCT.     

A fuzzy classifier with ellipsoidal regions for diagnosis problems

In our previous work, we developed a fuzzy classifier with ellipsoidal regions that has a training capability. In this paper, we extend the fuzzy classifier to diagnosis problems, in which the training data belonging to abnormal classes are difficult to obtain while the training data belonging to normal classes are easily obtained. Assuming that there are no data belonging to abnormal classes, we first train the fuzzy classifier with only the data belonging to normal classes. We then introduce the threshold of the minimum-weighted distance from the centers of the clusters for the data belonging to normal classes. If the unknown data is within the threshold, we classify the data into normal classes and, if not, abnormal classes. The operator checks whether the diagnosis is correct. If the incoming data is classified into the same normal class both by the classifier and the operator, nothing is done. But if the input data is classified into the different normal classes by the classifier and the operator, or if the incoming data is classified into an abnormal class, but the operator classified it into a normal class, the slopes of the membership functions of the fuzzy rules are tuned. If the operator classifies the data into an abnormal class, the classifier is retrained adding the newly obtained data irrespective of the classifier's classification result. The online training is continued until a sufficient number of the data belonging to abnormal classes are obtained. Then the threshold is optimized using the data belonging to both normal and abnormal classes. We evaluate our method using the Fisher iris data, blood cell data, and thyroid data, assuming some of the classes are abnormal     

Classification of multispectral remote sensing data using aback-propagation neural network

The suitability of a back-propagation neural network for classification of multispectral image data is explored. A methodology is developed for selection of both training parameters and data sets for the training phase. A new technique is also developed to accelerate the learning phase. To benchmark the network, the results are compared to those obtained using three other algorithms: a statistical contextual technique, a supervised piecewise linear classifier, and an unsupervised multispectral clustering algorithm. All three techniques were applied to simulated and real satellite imagery. Results from the classification of both Monte Carlo simulation and real imagery are summarized     

Integration of spatial and spectral information by means of unsupervised extraction and classification for homogenous objects applied to multispectral and hyperspectral data

This paper presents a method of unsupervised enhancement of pixels homogeneity in a local neighborhood. This mechanism will enable an unsupervised contextual classification of multispectral data that integrates the spectral and spatial information producing results that are more meaningful to the human analyst. This unsupervised classifier is an unsupervised development of the well-known supervised extraction and classification for homogenous objects (ECHO) classifier. One of its main characteristics is that it simplifies the retrieval process of spatial structures. This development is specially relevant for the new generation of airborne and spaceborne sensors with high spatial resolution.     

A study on several machine-learning methods for classification of malignant and benign clustered microcalcifications

In this paper, we investigate several state-of-the-art machine-learning methods for automated classification of clustered microcalcifications (MCs). The classifier is part of a computer-aided diagnosis (CADx) scheme that is aimed to assisting radiologists in making more accurate diagnoses of breast cancer on mammograms. The methods we considered were: support vector machine (SVM), kernel Fisher discriminant (KFD), relevance vector machine (RVM), and committee machines (ensemble averaging and AdaBoost), of which most have been developed recently in statistical learning theory. We formulated differentiation of malignant from benign MCs as a supervised learning problem, and applied these learning methods to develop the classification algorithm. As input, these methods used image features automatically extracted from clustered MCs. We tested these methods using a database of 697 clinical mammograms from 386 cases, which included a wide spectrum of difficult-to-classify cases. We analyzed the distribution of the cases in this database using the multidimensional scaling technique, which reveals that in the feature space the malignant cases are not trivially separable from the benign ones. We used receiver operating characteristic (ROC) analysis to evaluate and to compare classification performance by the different methods. In addition, we also investigated how to combine information from multiple-view mammograms of the same case so that the best decision can be made by a classifier. In our experiments, the kernel-based methods (i.e., SVM, KFD, and RVM) yielded the best performance (Az = 0.85, SVM), significantly outperforming a well-established, clinically-proven CADx approach that is based on neural network (Az = 0.80).     

A Gaussian adaptive resonance theory neural network classificationalgorithm applied to supervised land cover mapping using multitemporalvegetation index data

Neural network classifiers have been shown to provide supervised classification results that significantly improve on traditional classification algorithms such as the Bayesian (maximum likelihood [ML]) classifier. While the predominant neural network architecture has been the feedforward multilayer perceptron known as backpropagation. Adaptive resonance theory (ART) neural networks offer advantages to the classification of optical remote sensing data for vegetation and land cover mapping. A significant advantage is that it does not require prior specification of the neural net structure, creating as many internal nodes as are needed to represent the calibration (training) data. The Gaussian ARTMAP classification algorithm bases the probability that input training samples belong to specific classes on the parameters of its Gaussian distributions: the means, standard deviations, and a priori probabilities. The performance of the Gaussian ARTMAP classification algorithm in terms of classification accuracy using independent validation data indicated was over 70% accurate when applied to an annual series of monthly 1-km advanced very high resolution radiometer (AVHRR) satellite normalized difference vegetation index (NDVI) data. The accuracies were comparable to those of fuzzy ARTMAP and a univariate decision tree, and significantly higher than a Bayesian classification algorithm. Algorithm testing is based on calibration and validation data developed and applied to Central America to map the International Geosphere-Biosphere Programme (IGBP) land cover classification system     

一种PMN网络的GEM训练算法

本文提出一种概率映射网络的ＧＥＭ训练算法，它是ＥＭ算法的一种改进算法。ＰＭＮ风为一个四层前馈网。它构成一个贝叶斯分类器，实验多类分类的贝叶斯判别，把输入的样本模式经网络变换为输出的分类判决，其网络节点对应于贝叶斯后验概率公式的各个变量。此ＰＭＮ网络用高斯核函数作为密度函数，网络参数的训练由ＧＥＭ算法实现，其学习方式为类间的监督学习和类内的非监督学习。