Neural network-based segmentation and classification system for automated grading of histologic sections of bladder carcinoma

OBJECTIVE: To develop an image analysis system for automated nuclear segmentation and classification of histologic bladder sections employing quantitative nuclear features. STUDY DESIGN: Ninety-two cases were classified into three classes by experienced pathologists according to the WHO grading system: 18 cases as grade 1, 45 as grade 2, and 29 as grade 3. Nuclear segmentation was performed by means of an artificial neural network (ANN)-based pixel classification algorithm, and each case was represented by 36 nuclei features. Automated grading of bladder tumor histologic sections was performed by an ANN classifier implemented in a two-stage hierarchic tree. RESULTS: On average, 95% of the nuclei were correctly detected. At the first stage of the hierarchic tree, classifier performance in discriminating between cases of grade 1 and 2 and cases of grade 3 was 89%. At the second stage, 79% of grade 1 cases were correctly distinguished from grade 2 cases. CONCLUSION: The proposed image analysis system provides the means to reduce subjectivity in grading bladder tumors and may contribute to more accurate diagnosis and prognosis since it relies on nuclear features, the value of which has been confirmed.     

Computer-based malignancy grading of astrocytomas employing a support vector machine classifier, the WHO grading system and the regular hematoxylin-eosin diagnostic staining procedure

OBJECTIVE: To investigate and develop an automated technique for astrocytoma malignancy grading compatible with the clinical routine. STUDY DESIGN: One hundred forty biopsies of astrocytomas were collected from 2 hospitals. The degree of tumor malignancy was defined as low or high according to the World Health Organization grading system. From each biopsy, images were digitized and segmented to isolate nuclei from background tissue. Morphologic and textural nuclear features were quantified to encode tumor malignancy. Each case was represented by a 40-dimensional feature vector. An exhaustive search procedure in feature space was utilized to determine the best feature combination that resulted in the smallest classification error. Low and high grade tumors were discriminated using support vector machines (SVMs). To evaluate the system performance, all available data were split randomly into training and test sets. RESULTS: The best vector combination consisted of 3 textural and 2 morphologic features. Low and high grade cases were discriminated with an accuracy of 90.7% and 88.9%, respectively, using an SVM classifier with polynomial kernel of degree 2. CONCLUSION: The proposed methodology was based on standards that are common in daily clinical practice and might be used in parallel with conventional grading as a second-opinion tool to reduce subjectivity in the classification of astrocytomas.     

Topometric analysis of diffuse astrocytomas

OBJECTIVE: To investigate differences between astrocytomas of WHO grade 2 and anaplastic astrocytomas of WHO grade 3 in terms of topometric variables characterizing individual tumor cell nuclei. STUDY DESIGN: Paraffin sections from surgical specimens from 25 astrocytomas (grade 2, n = 11; grade 3, n = 14) were analyzed by means of an image analysis system. At least 300 tumor cell nuclei were measured in the region with the highest Ki-67 proliferation index. Three different kinds of topometric variables were determined for each tumor cell nucleus: (1) several distances; (2) the variable Angle 2/1, the angle between the straight lines representing the distance to the nearest and second-nearest nucleus; and (3) the number of neighbors according to our mathematical definition. RESULTS: Most topometric variables showed distinct differences between the 2 tumor grades (multivariate analysis of variance), with 88% cases correctly reclassified by means of cross-validated discriminant analysis. The variables with the highest discriminatory power were the SD of Angle 2/1 and the ratio between the distance to the second-nearest and nearest tumor cell nucleus, with lower values for these variables in anaplastic astrocytomas. Even variables concerning neighborhood relationships showed significant differences. CONCLUSION: The results of this pilot study show that this first set of topometric variables is sufficient to detect differences between topologic characteristics of tumor cell nuclei in astrocytomas grade 2 and grade 3. Topometry seems to be an important tool for grading astrocytomas.     

Heart rate variability dynamics for the prognosis of cardiovascular risk

Statistical, spectral, multi-resolution and non-linear methods were applied to heart rate variability (HRV) series linked with classification schemes for the prognosis of cardiovascular risk. A total of 90 HRV records were analyzed: 45 from healthy subjects and 45 from cardiovascular risk patients. A total of 52 features from all the analysis methods were evaluated using standard two-sample Kolmogorov-Smirnov test (KS-test). The results of the statistical procedure provided input to multi-layer perceptron (MLP) neural networks, radial basis function (RBF) neural networks and support vector machines (SVM) for data classification. These schemes showed high performances with both training and test sets and many combinations of features (with a maximum accuracy of 96.67%). Additionally, there was a strong consideration for breathing frequency as a relevant feature in the HRV analysis.     

Genetic analysis to complement histopathological diagnosis of brain tumors

Gliomas, the most frequent tumors originating in the human nervous system, are divided into various subtypes. Currently, microscopic examination alone is insufficient for classification and grading so that genetic profiles are increasingly being emphasized in recognition of the emerging role of molecular diagnostic approaches to glioma classification. Glioblastomas (WHO grade IV) may develop de novo (primary glioblastomas) or through progression from lower-grade astrocytomas (secondary glioblastomas), while both glioblastomas show similar histological features. In contrast, they do constitute distinct disease entities that evolve through different genetic pathways, and are likely to differ in prognosis and response to therapy. Oligodendrogliomas (WHO grade II) account for 2.7% of brain tumors and 5-18% of all gliomas. Since this tumor is recognized as a particular subtype of glioma that shows remarkable responses to chemotherapy, a correct diagnosis is of prime importance. The difficulty is that histological differentiation of oligodendrogliomas from diffuse astrocytomas is highly subjective in cases without typical morphological features and there is a lack of reliable immunohistochemical markers. While histological distinction of low-grade gliomas from reactive astrocytes is also often difficult, reactive astrocytes usually lack genetic alterations. More biological and molecular approaches to glioma classification thus appear warranted to provide improved means to achieve correct diagnoses.     

Diagnosis of human oligodendrogliomas with the help of the NeuroShell Easy Classifier neural network

OBJECTIVE: To examine whether a suitable solution can be found concerning the ability to reproduce the histologic classification of human oligodendrogliomas with the assistance of the NeuroShell Easy Classifier neural network. STUDY DESIGN: Histologic sections of 449 human oligodendrogliomas were selected. The diagnostic task was given by differentiation of three oligodendroglioma types: 121 low grade oligodendrogliomas, World Health Organization grade 2; 180 low grade oligoastrocytomas; and 148 anaplastic oligodendrogliomas, grade 3. Age, sex and 50 histologic characteristics were examined in each case, describing the presence of a specific histologic feature on a scale of four (zero, absence of the feature; three, abundant presence). From each group, two-thirds of randomly selected tumors were available for the training set and one-third for the testing set. RESULTS: In the three-class problem, 98.88% of the tumors were correctly classified (testing set). Ninety-nine percent of new testing tumors were correctly classified with Easy Classifier as low grade and anaplastic oligodendrogliomas. In the case of low grade oligodendrogliomas versus low grade oligoastrocytomas, 99% of new tumors were correctly classified. CONCLUSION: The main conclusion from this study is that Easy Classifier was able to differentiate, with high accuracy, sensitivity and specificity, among the three types of oligodendrogliomas.     

Exploring Radiologic Criteria for Glioma Grade Classification on the BraTS Dataset

ObjectivesGlioma grading using maching learning on magnetic resonance data is a growing topic. According to the World Health Organization (WHO), the classification of glioma discriminates between low grade gliomas (LGG), grades I, II; and high grade gliomas (HGG), grades III, IV, leading to major issues in oncology for therapeutic management of patients. A well-known dataset for machine-based grade prediction is the MICCAI Brain Tumor Segmentation (BraTS) dataset. However this dataset is not divided into WHO-defined LGG and HGG, since it combines grades I, II and III as “lower grades gliomas”, while its HGG category only presents grade IV glioblastoma multiform. In this paper we want to train a binary grade classifier and investigate the consistency of the original BraTS labels with radiologic criteria using machine-aided predictions.Material and methodsUsing WHO-based radiomic features, we trained a SVM classifier on the BraTS dataset, and used the prediction score histogram to investigate the behaviour of our classifier on the lower grade population. We also asked 5 expert radiologists to annotate BraTS images between low (as opposed to lower) grade and high grade glioma classes, resulting in a new groundtruth.ResultsOur first training reached 84.1% accuracy. The prediction score histogram allows us to identify the radiologically high grade patients among the original lower grade population of the BraTS dataset. Training another SVM on our new radiologically WHO-aligned groundtruth shows robust performances despite important class imbalance, reaching 82.4% accuracy.ConclusionOur results highlight the coherence of radiologic criteria for low grade versus high grade classification under WHO terms. We also show how the histogram of prediction scores and crossed prediction scores can be used as tools for data exploration and performance evaluation. Therefore, we propose to use our radiological groundtruth for future development on binary glioma grading.     

Prediction of ATP/GTP-binding motif: a comparison of a perceptron type neural network and a consensus sequence method [corrected

Neural networks have been applied to a number of protein structure problems. In some applications their success has not been substantiated by a comparison with the performance of a suitable alternative statistical method on the same data. In this paper, a two-layer feed-forward neural network has been trained to recognize ATP/GTP-binding [corrected] local sequence motifs. The neural network correctly classified 78% of the 349 sequences used. This was much better than a simple motif-searching program. A more sophisticated statistical method was developed, however, which performed marginally better (80% correct classification) than the neural network. The neural network and the statistical method performed similarly on sequences of varying degrees of homology. These results do not imply that neural networks, especially those with hidden layers, are not useful tools, but they do suggest that two-layer networks in particular should be carefully tested against other statistical methods.     

AllerTool: a web server for predicting allergenicity and allergic cross-reactivity in proteins

Assessment of potential allergenicity and patterns of cross-reactivity is necessary whenever novel proteins are introduced into human food chain. Current bioinformatic methods in allergology focus mainly on the prediction of allergenic proteins, with no information on cross-reactivity patterns among known allergens. In this study, we present AllerTool, a web server with essential tools for the assessment of predicted as well as published cross-reactivity patterns of allergens. The analysis tools include graphical representation of allergen cross-reactivity information; a local sequence comparison tool that displays information of known cross-reactive allergens; a sequence similarity search tool for assessment of cross-reactivity in accordance to FAO/WHO Codex alimentarius guidelines; and a method based on support vector machine (SVM). A 10-fold cross-validation results showed that the area under the receiver operating curve (A(ROC)) of SVM models is 0.90 with 86.00% sensitivity (SE) at specificity (SP) of 86.00%. Availability: AllerTool is freely available at http://research.i2r.a-star.edu.sg/AllerTool/.     

Detection of dairy fouling: Combining ultrasonic measurements and classification methods

Fouling and cleaning in heat exchangers are severe and costly (up to 0.3% of gross national product) issues in dairy and food processing. Therefore, reducing cleaning time and cost is urgently needed. In this study, two classification methods [artificial neural network (ANN) and support vector machine (SVM)] for detecting protein and mineral fouling presence and absence based on ultrasonic measurements were presented and compared. ANN is based on a multilayer perceptron feed forward neural network, whereas SVM is based on clustering between fouling and no fouling using a hyperplane. When both fouling types (1239 datasets) were combined, ANN showed an accuracy of 71.9% while SVM displayed an accuracy of 97.6%. Separate fouling detection of mineral/protein fouling by ANN/SVM was comparable: dependent on fouling type detection accuracies of 100% (protein fouling, ANN and SVM), and 98.2% (SVM), and 93.5% (ANN) for mineral fouling was reached. It was shown that it was possible to detect fouling presence and absence offline in a static setup using ultrasonic measurements in combination with a classification method. This study proved the applicability of combining classification methods and fouling measurements to take a step toward reducing cleaning costs and time.     

Application of different methods for nuclear shape analysis with special reference to the differentiation of brain tumors

OBJECTIVE: To study the discriminatory power of different methods designed for nuclear shape analysis with reference to the differentiation and grading of brain tumors and the differentiation between proliferating and nonproliferating nuclei. STUDY DESIGN: At least 300 tumor cell nuclei per case were measured by means of a digital image analysis system. Fourier amplitudes no. 1 to 15, moments no. 1 to 7 according to Hu, roundness factor, ellipse shape factor, concavity factor, Feret ratio, fractal dimension and bending energy were determined for each nucleus. The discriminatory power of these parameters was tested in three pairwise comparisons: (1) oligodendrogliomas WHO grade II (n = 13) vs. grade III (n = 11), (2) medulloblastomas WHO grade IV (n = 14) vs. anaplastic ependymomas WHO grade III (n = 12), (3) Ki-67-positive vs. Ki-67-negative tumor cell nuclei in the 14 medulloblastomas. RESULTS: When data from Fourier analysis were included in statistical analysis, cross-validated discriminant analysis led to a 100% correct reclassification for the first and for the second pairwise comparison and to a 75% correct reclassification when comparing Ki-67-positive and Ki-67-negative nucleifrom medulloblastomas. Different combinations of the other shape parameters led to a lower percentage of correctly reclassified cases for all three pairwise comparisons, especially when Fourier analysis was not included in the analysis. CONCLUSION: Fourier analysis provided an optimal statistical discrimination between different brain tumor entities and between data sets from proliferating and nonproliferating tumor cell nuclei. Since nuclear shape is an important criterion for the investigation of tumors, the application of Fourier analysis is therefore recommended for quantitative histologic investigations in neuro-oncology.     

Enzyme family classification by support vector machines

One approach for facilitating protein function prediction is to classify proteins into functional families. Recent studies on the classification of G-protein coupled receptors and other proteins suggest that a statistical learning method, Support vector machines (SVM), may be potentially useful for protein classification into functional families. In this work, SVM is applied and tested on the classification of enzymes into functional families defined by the Enzyme Nomenclature Committee of IUBMB. SVM classification system for each family is trained from representative enzymes of that family and seed proteins of Pfam curated protein families. The classification accuracy for enzymes from 46 families and for non-enzymes is in the range of 50.0% to 95.7% and 79.0% to 100% respectively. The corresponding Matthews correlation coefficient is in the range of 54.1% to 96.1%. Moreover, 80.3% of the 8,291 correctly classified enzymes are uniquely classified into a specific enzyme family by using a scoring function, indicating that SVM may have certain level of unique prediction capability. Testing results also suggest that SVM in some cases is capable of classification of distantly related enzymes and homologous enzymes of different functions. Effort is being made to use a more comprehensive set of enzymes as training sets and to incorporate multi-class SVM classification systems to further enhance the unique prediction accuracy. Our results suggest the potential of SVM for enzyme family classification and for facilitating protein function prediction. Our software is accessible at http://jing.cz3.nus.edu.sg/cgi-bin/svmprot.cgi.     

Classification across gene expression microarray studies

Background  

The increasing number of gene expression microarray studies represents an important resource in biomedical research. As a result, gene expression based diagnosis has entered clinical practice for patient stratification in breast cancer. However, the integration and combined analysis of microarray studies remains still a challenge. We assessed the potential benefit of data integration on the classification accuracy and systematically evaluated the generalization performance of selected methods on four breast cancer studies comprising almost 1000 independent samples. To this end, we introduced an evaluation framework which aims to establish good statistical practice and a graphical way to monitor differences. The classification goal was to correctly predict estrogen receptor status (negative/positive) and histological grade (low/high) of each tumor sample in an independent study which was not used for the training. For the classification we chose support vector machines (SVM), predictive analysis of microarrays (PAM), random forest (RF) and k-top scoring pairs (kTSP). Guided by considerations relevant for classification across studies we developed a generalization of kTSP which we evaluated in addition. Our derived version (DV) aims to improve the robustness of the intrinsic invariance of kTSP with respect to technologies and preprocessing.     

Predicting CNS permeability of drug molecules: comparison of neural network and support vector machine algorithms.

Two different machine-learning algorithms have been used to predict the blood-brain barrier permeability of different classes of molecules, to develop a method to predict the ability of drug compounds to penetrate the CNS. The first algorithm is based on a multilayer perceptron neural network and the second algorithm uses a support vector machine. Both algorithms are trained on an identical data set consisting of 179 CNS active molecules and 145 CNS inactive molecules. The training parameters include molecular weight, lipophilicity, hydrogen bonding, and other variables that govern the ability of a molecule to diffuse through a membrane. The results show that the support vector machine outperforms the neural network. Based on over 30 different validation sets, the SVM can predict up to 96% of the molecules correctly, averaging 81.5% over 30 test sets, which comprised of equal numbers of CNS positive and negative molecules. This is quite favorable when compared with the neural network's average performance of 75.7% with the same 30 test sets. The results of the SVM algorithm are very encouraging and suggest that a classification tool like this one will prove to be a valuable prediction approach.     

Identification of COL6A1 as a differentially expressed gene in human astrocytomas

Diffuse infiltrating gliomas are the most common tumors of the central nervous system. Gliomas are classified by the WHO according to their histopathological and clinical characteristics into four classes: grade I (pilocytic astrocytoma), grade II (diffuse astrocytoma), grade III (anaplastic astrocytoma), and grade IV (glioblastoma multiforme). Several genes have already been correlated with astrocytomas, but many others are yet to be uncovered. By analyzing the public SAGE data from 21 patients, comprising low malignant grade astrocytomas and glioblastomas, we found COL6A1 to be differentially expressed, confirming this finding by real time RT-PCR in 66 surgical samples. To the best of our knowledge, COL6A1 has never been described in gliomas. The expression of this gene has significantly different means when normal glia is compared with low-grade astrocytomas (grades I and II) and high-grade astrocytomas (grades III and IV), with a tendency to be greater in higher grade samples, thus rendering it a powerful tumor marker.     

一种基于支持向量机技术的癫痫脑电棘尖波识别方法

支持向量机是一种基于统计学习理论的新型学习机。文章提出一种基于支持向量机的癫痫脑电特征提取与识别方法,充分发挥其泛化能力强的特点,在与神经网络方法的比较中,表现出较低的漏检率和较好的鲁棒性,有深入研究的价值和良好的应用前景。     

Computer-aided classification of malignancy in astrocytomas. I. The value of nuclear parameters obtained by automated black and white image analysis.

Ninety-three selected cases of astrocytomas including glioblastomas (astrocytomas grades 1-4) were evaluated by means of Feulgen-stained microscopic slides for nuclear parameters obtained by automated black and white image analysis (ABWIA). The goal was to determine to what extent nuclear features evaluated by ABWIA were applicable as classifiers for the computer-aided numerical classification of malignancy in astrocytomas. Before the automated evaluation, all tumours had been subjectively graded according to the Mayo Clinic grading rules as delineated by Ringertz. Twenty-three nuclear parameters were evaluated and tested for their classification impact. With a model of five parameters (number of nuclei per area, mean of the convex form factor, extinction sum, extinction variation, and full-width-half-maximum of the extinction distribution) the highest reclassification rate of 75% correctly reclassified cases was obtained. Although this is a good result for a classification using only nuclear parameters, it is too poor for practical application. Thus, nuclear parameters evaluated by ABWIA alone are insufficient for numerical classification models assessing the malignant expression of astrocytomas.     

Morphometry, densitometry and pattern analysis of plastic-embedded histologic material from urothelial cell carcinoma of the bladder.

An image analysis method of grading histologic sections of bladder carcinoma was tested. The method was new in four respects. First, for fixation of the biopsies a coagulant fixative was used. Second, 2-microns plastic sections were used to ensure the reproducibility of nuclear imaging. Third, a new stereologic approach was used for calculation of the nuclear volume and DNA content. Fourth, for the classification rule the morphometric, densitometric and texture features were used in concert. The IBAS 2000 instrument was used for the measurements. Texture analysis of the chromatin patterns was performed using Markovian texture features. Using discriminant analysis, of 22 parameters, 2 morphometric, 2 densitometric and 3 texture features were selected for the classification rule. With them, 89% of the bladder carcinomas were correctly classified into the three grades. All grade III tumors were classified correctly. Among the features tested, the densitometry of the DNA had the highest F values. All of the grade III tumors and 45% of the grade II tumor group had DNA histograms indicating aneuploidy. This study showed that plastic-embedded material is well suited to morphometry and densitometry and can be used for quantitative grading of bladder carcinoma.