Bagging tree classifiers for laser scanning images: a data- and simulation-based strategy

Application of nanotechnology in breast cancer screening under obstetrics and gynecology through the use of CNN and ANFIS

Breast cancer is the leading reason of death among women aged 35 to 54. Breast cancer diagnosis still presents significant challenges, and preventing the disease's most severe symptoms requires early detection. The role of nanotechnology in the tumor-treatment has recently attracted a lot of interest. In cancer therapies, nanotechnology plays a major role in the medication distribution process. Nanoparticles have the ability to target tumors. Nanoparticles are favorable and maybe preferable for usage in tumor detection and imaging due to their incredibly small size. Quantum dots, semiconductor crystals with increased labeling and imaging capabilities for cancer cells, are one of the particles that have received the most research attention. The design of the research is cross-sectional and descriptive. From April through September of 2020, data were gathered at the State Hospital. All pregnant women who came to the hospital throughout the first and second trimesters of the research's data collection were included in the study population. 100 pregnant women between the ages of 20 and 40 who had not yet had a mammogram comprised the research sample. 1100 digitized mammography images are included in the dataset, which was obtained from a hospital. Convolutional neural networks (CNN) were used to scan all images, and breast masses and mass comparisons were made using the malignant-benign categorization. The adaptive neuro-fuzzy inference system (ANFIS) then examined all of the data obtained by CNN in order to identify breast cancer early using inputs based on the nine different inputs. The precision of the mechanism used in this technique to determine the ideal radius value is significantly impacted by the radius value. Nine variables that define breast cancer indicators were utilized as inputs to the ANFIS classifier, which was then used to identify breast cancer. The parameters were given the necessary fuzzy functions, and the combined dataset was applied to train the method. Testing was initially performed by 30% of dataset that was later done with the real data obtained from the hospital. The accuracy of the results for 30% data was 84% (specificity =72.7%, sensitivity =86.7%) and the results for the real data was 89.8% (sensitivity =82.3%, specificity =75.9%), respectively.

Decision trees: from efficient prediction to responsible AI

This article provides a birds-eye view on the role of decision trees in machine learning and data science over roughly four decades. It sketches the evolution of decision tree research over the years, describes the broader context in which the research is situated, and summarizes strengths and weaknesses of decision trees in this context. The main goal of the article is to clarify the broad relevance to machine learning and artificial intelligence, both practical and theoretical, that decision trees still have today.

CCPNet136: automated detection of schizophrenia using carbon chain pattern and iterative TQWT technique with EEG signals

Objective.Schizophrenia (SZ) is a severe, chronic psychiatric-cognitive disorder. The primary objective of this work is to present a handcrafted model using state-of-the-art technique to detect SZ accurately with EEG signals.Approach.In our proposed work, the features are generated using a histogram-based generator and an iterative decomposition model. The graph-based molecular structure of the carbon chain is employed to generate low-level features. Hence, the developed feature generation model is called the carbon chain pattern (CCP). An iterative tunable q-factor wavelet transform (ITQWT) technique is implemented in the feature extraction phase to generate various sub-bands of the EEG signal. The CCP was applied to the generated sub-bands to obtain several feature vectors. The clinically significant features were selected using iterative neighborhood component analysis (INCA). The selected features were then classified using the k nearest neighbor (kNN) with a 10-fold cross-validation strategy. Finally, the iterative weighted majority method was used to obtain the results in multiple channels.Main results.The presented CCP-ITQWT and INCA-based automated model achieved an accuracy of 95.84% and 99.20% using a single channel and majority voting method, respectively with kNN classifier.Significance.Our results highlight the success of the proposed CCP-ITQWT and INCA-based model in the automated detection of SZ using EEG signals.

Fibrosis of Peritoneal Membrane, Molecular Indicators of Aging and Frailty Unveil Vulnerable Patients in Long-Term Peritoneal Dialysis

Peritoneal membrane status, clinical data and aging-related molecules were investigated as predictors of long-term peritoneal dialysis (PD) outcomes. A 5-year prospective study was conducted with the following endpoints: (a) PD failure and time until PD failure, (b) major cardiovascular event (MACE) and time until MACE. A total of 58 incident patients with peritoneal biopsy at study baseline were included. Peritoneal membrane histomorphology and aging-related indicators were assessed before the start of PD and investigated as predictors of study endpoints. Fibrosis of the peritoneal membrane was associated with MACE occurrence and earlier MACE, but not with the patient or membrane survival. Serum α-Klotho bellow 742 pg/mL was related to the submesothelial thickness of the peritoneal membrane. This cutoff stratified the patients according to the risk of MACE and time until MACE. Uremic levels of galectin-3 were associated with PD failure and time until PD failure. This work unveils peritoneal membrane fibrosis as a window to the vulnerability of the cardiovascular system, whose mechanisms and links to biological aging need to be better investigated. Galectin-3 and α-Klotho are putative tools to tailor patient management in this home-based renal replacement therapy.

Developing clinical algorithm for identifying acute lumbar spondylolysis in elementary school children - Classification and regression tree analysis

OBJECTIVES

To develop a clinical algorithm for classifying acute lumbar spondylolysis from nonspecific low back pain in elementary school-aged patients using the classification and regression tree analysis.

METHODS

Medical records of 73 school-aged patients diagnosed with acute lumbar spondylolysis or nonspecific low back pain were retrospectively reviewed. Fifty-eight patients were examined for establishing an algorithm and 15 were employed for testing its performance. The following data were retrieved: age, gender, school grades, days after symptom onset, history of low back pain, days of past low back pain, height, weight, body mass index, passive straight leg raise test results, hours per week spent on sports activities, existence of spina bifida, lumbar lordosis angle, and lumbosacral joint angle. Classification and regression tree analyses were performed 150 times using the bootstrap and aggregating method. Then, the results were integrated by majority vote, establishing an algorithm.

RESULTS

Lumbar lordosis angle, days after symptom onset, body mass index, and lumbosacral joint angle were the predictors for classifying those injuries.

CONCLUSION

The algorithm can be used to identify elementary school-aged children with low back pain requiring advanced imaging investigation, although a future study with a larger sample population is necessary for validating the algorithm.

Novel favipiravir pattern-based learning model for automated detection of specific language impairment disorder using vowels

Specific language impairment (SLI) is one of the most common diseases in children, and early diagnosis can help to obtain better timely therapy economically. It is difficult and time-consuming for clinicians to accurately detect SLI through standard clinical assessments. Hence, machine learning algorithms have been developed to assist in the accurate diagnosis of SLI. This work aims to investigate the graph of the favipiravir molecule-based feature extraction function and propose an accurate SLI detection model using vowels. We proposed a novel handcrafted machine learning framework. This architecture comprises the favipiravir molecular structure pattern, statistical feature extractor, wavelet packet decomposition (WPD), iterative neighborhood component analysis (INCA), and support vector machine (SVM) classifier. Two feature extraction models, statistical and textural, are employed in the handcrafted feature generation methodology. A new nature-inspired graph-based feature extractor that uses the chemical depiction of the favipiravir (favipiravir became popular with the COVID-19 pandemic) is employed for feature extraction. Finally, the proposed favipiravir pattern, statistical feature extractor, and wavelet packet decomposition are used to create a feature vector. Moreover, a statistical feature extractor is used in this work. The WPD generates multilevel features, and the most meaningful features are selected using the NCA feature selector. Finally, these chosen features are fed to SVM classifier for automated classification. Two validation methods, (i) leave one subject out (LOSO) and (ii) tenfold cross-validations (CV), are used to obtain robust classification results. Our proposed favipiravir pattern-based model developed using a vowel dataset can detect SLI children with an accuracy of 99.87% and 98.86% using tenfold and LOSO CV strategies, respectively. These results demonstrated the high vowel classification ability of the proposed favipiravir pattern-based model.

Hematopoietic Prostaglandin D2 Synthase Controls Tfh/Th2 Communication and Limits Tfh Antitumor Effects

T follicular helper (Tfh) cells are a subset of CD4+ T cells essential in immunity and have a role in helping B cells produce antibodies against pathogens. However, their role during cancer progression remains unknown. The mechanism of action of Tfh cells remains elusive because contradictory data have been reported on their protumor or antitumor responses in human and murine tumors. Like Tfh cells, Th2 cells are also involved in humoral immunity and are regularly associated with tumor progression and poor prognosis, mainly through their secretion of IL4. Here, we showed that Tfh cells expressed hematopoietic prostaglandin D2 (PGD2) synthase in a pSTAT1/pSTAT3-dependent manner. Tfh cells produced PGD2, which led to recruitment of Th2 cells via the PGD2 receptor chemoattractant receptor homologous molecule expressed on Th type 2 cells (CRTH2) and increased their effector functions. This cross-talk between Tfh and Th2 cells promoted IL4-dependent tumor growth. Correlation between Th2 cells, Tfh cells, and hematopoietic PGD2 synthase was observed in different human cancers and associated with outcome. This study provides evidence that Tfh/Th2 cross-talk through PGD2 limits the antitumor effects of Tfh cells and, therefore, could serve as a therapeutic target.

Deep Learning Image Analysis of Optical Coherence Tomography Angiography Measured Vessel Density Improves Classification of Healthy and Glaucoma Eyes

PURPOSE

To compare convolutional neural network (CNN) analysis of en face vessel density images to gradient boosting classifier (GBC) analysis of instrument-provided, feature-based optical coherence tomography angiography (OCTA) vessel density measurements and OCT retinal nerve fiber layer (RNFL) thickness measurements for classifying healthy and glaucomatous eyes.

DESIGN

Comparison of diagnostic approaches.

METHODS

A total of 130 eyes of 80 healthy individuals and 275 eyes of 185 glaucoma patients with optic nerve head (ONH) OCTA and OCT imaging were included. Classification performance of a VGG16 CNN trained and tested on entire en face 4.5 × 4.5-mm radial peripapillary capillary OCTA ONH images was compared to the performance of separate GBC models trained and tested on standard OCTA and OCT measurements. Five-fold cross-validation was used to test predictions for CNNs and GBCs. Areas under the precision recall curves (AUPRC) were calculated to control for training/test set size imbalance and were compared.

RESULTS

Adjusted AUPRCs for GBC models were 0.89 (95% CI = 0.82, 0.92) for whole image vessel density GBC, 0.89 (0.83, 0.92) for whole image capillary density GBC, 0.91 (0.88, 0.93) for combined whole image vessel and whole image capillary density GBC, and 0.93 (0.91, 095) for RNFL thickness GBC. The adjusted AUPRC using CNN analysis of en face vessel density images was 0.97 (0.95, 0.99) resulting in significantly improved classification compared to GBC OCTA-based results and GBC OCT-based results (P ≤ 0.01 for all comparisons).

CONCLUSION

Deep learning en face image analysis improves on feature-based GBC models for classifying healthy and glaucoma eyes.

Prediction of prognosis and survival of patients with gastric cancer by a weighted improved random forest model: an application of machine learning in medicine

INTRODUCTION

It is essential to predict the survival status of patients based on their prognosis. This can assist physicians in evaluating treatment decisions. Random forest is an excellent machine learning algorithm even without any modification. We propose a new random forest weighting method and apply it to the gastric cancer patient data from the Surveillance, Epidemiology, and End Results (SEER) program. We evaluated the generalization ability of this weighted random forest algorithm on 10 public medical datasets. Furthermore, for the same weighting mode, the difference between using out-of-bag (OOB) data and all training sets as the weighting basis is explored.

MATERIAL AND METHODS

110 697 cases of gastric cancer patients diagnosed between 1975 and 2016 obtained from the SEER database were included in the experiment. In addition, 10 public medical datasets were used for the generalization ability evaluation of this weighted random forest algorithm.

RESULTS

Through experimental verification, on the SEER gastric cancer patient data, the weighted random forest algorithm improves the accuracy by 0.79% compared with the original random forest. In AUC, macro-averaging increased by 2.32% and micro-averaging increased by 0.51% on average. Among the 10 public datasets, the random forest weighted in accuracy has the best performance on 6 datasets, with an average increase of 1.44% in accuracy and an average increase of 1.2% in AUC.

CONCLUSIONS

Compared with the original random forest, the weighted random forest model shows a significant improvement in performance, and the effect of using all training data as the weighting basis is better than using OOB data.

Coronary Angiography Print: An Automated Accurate Hidden Biometric Method Based on Filtered Local Binary Pattern Using Coronary Angiography Images

Background and purpose: Biometrics is a commonly studied research issue for both biomedical engineering and forensics sciences. Besides, the purpose of hidden biometrics is to discover hidden biometrics features. This work aims to demonstrate the biometric identification ability of coronary angiography images. Material and method: A new coronary angiography images database was collected to develop an automatic identification model. The used database was collected from 51 subjects and contains 2156 images. The developed model has to preprocess; feature generation using local binary pattern; feature selection with neighborhood component analysis; and classification phases. In the preprocessing phase; image rotations; median filter; Gaussian filter; and speckle noise addition functions have been used to generate filtered images. A multileveled extractor is presented using local binary pattern and maximum pooling together. The generated features are fed to neighborhood component analysis and the selected features are classified using k nearest neighbor classifier. Results: The presented angiography image identification method attained 99.86% classification accuracy on the collected database. Conclusions: The obtained findings demonstrate that the angiography images can be utilized as biometric identification. Moreover, we discover a new hidden biometric feature using coronary angiography images and name of this hidden biometric is coronary angiography print.

PrimePatNet87: Prime pattern and tunable q-factor wavelet transform techniques for automated accurate EEG emotion recognition

Nowadays, many deep models have been presented to recognize emotions using electroencephalogram (EEG) signals. These deep models are computationally intensive, it takes a longer time to train the model. Also, it is difficult to achieve high classification performance using for emotion classification using machine learning techniques. To overcome these limitations, we present a hand-crafted conventional EEG emotion classification network. In this work, we have used novel prime pattern and tunable q-factor wavelet transform (TQWT) techniques to develop an automated model to classify human emotions. Our proposed cognitive model comprises feature extraction, feature selection, and classification steps. We have used TQWT on the EEG signals to obtain the sub-bands. The prime pattern and statistical feature generator are employed on the generated sub-bands and original signal to generate 798 features. 399 (half of them) out of 798 features are selected using minimum redundancy maximum relevance (mRMR) selector, and misclassification rates of each signal are evaluated using support vector machine (SVM) classifier. The proposed network generated 87 feature vectors hence, this model is named PrimePatNet87. In the last step of the feature generation, the best 20 feature vectors which are selected based on the calculated misclassification rates, are concatenated. The generated feature vector is subjected to the feature selection and the most significant 1000 features are selected using the mRMR selector. These selected features are then classified using an SVM classifier. In the last phase, iterative majority voting has been used to generate a general result. We have used three publicly available datasets, namely DEAP, DREAMER, and GAMEEMO, to develop our proposed model. Our presented PrimePatNet87 model reached over 99% classification accuracy on whole datasets with leave one subject out (LOSO) validation. Our results demonstrate that the developed prime pattern network is accurate and ready for real-world applications.

Feed-forward LPQNet based Automatic Alzheimer's Disease Detection Model

BACKGROUND

Alzheimer's disease (AD) is one of the most commonly seen brain ailments worldwide. Therefore, many researches have been presented about AD detection and cure. In addition, machine learning models have also been proposed to detect AD promptly.

MATERIALS AND METHOD

In this work, a new brain image dataset was collected. This dataset contains two categories, and these categories are healthy and AD. This dataset was collected from 1070 subjects. This work presents an automatic AD detection model to detect AD using brain images automatically. The presented model is called a feed-forward local phase quantization network (LPQNet). LPQNet consists of (i) multilevel feature generation based on LPQ and average pooling, (ii) feature selection using neighborhood component analysis (NCA), and (iii) classification phases. The prime objective of the presented LPQNet is to reach high accuracy with low computational complexity. LPQNet generates features on six levels. Therefore, 256 × 6 = 1536 features are generated from an image, and the most important 256 out 1536 features are selected. The selected 256 features are classified on the conventional classifiers to denote the classification capability of the generated and selected features by LPQNet.

RESULTS

The presented LPQNet was tested on three image datasets to demonstrate the universal classification ability of the LPQNet. The proposed LPQNet attained 99.68%, 100%, and 99.64% classification accuracy on the collected AD image dataset, the Harvard Brain Atlas AD dataset, and the Kaggle AD dataset. Moreover, LPQNet attained 99.62% accuracy on the Kaggle AD dataset using four classes.

CONCLUSIONS

Moreover, the calculated results from LPQNet are compared to other automatic AD detection models. Comparisons, results, and findings clearly denote the superiority of the presented model. In addition, a new intelligent AD detector application can be developed for use in magnetic resonance (MR) and computed tomography (CT) devices. By using the developed automated AD detector, new generation intelligence MR and CT devices can be developed.

High interpretable machine learning classifier for early glaucoma diagnosis

AIM

To develop a classifier for differentiating between healthy and early stage glaucoma eyes based on peripapillary retinal nerve fiber layer (RNFL) thicknesses measured with optical coherence tomography (OCT), using machine learning algorithms with a high interpretability.

METHODS

Ninety patients with early glaucoma and 85 healthy eyes were included. Early glaucoma eyes showed a visual field (VF) defect with mean deviation >-6.00 dB and characteristic glaucomatous morphology. RNFL thickness in every quadrant, clock-hour and average thickness were used to feed machine learning algorithms. Cluster analysis was conducted to detect and exclude outliers. Tree gradient boosting algorithms were used to calculate the importance of parameters on the classifier and to check the relation between their values and its impact on the classifier. Parameters with the lowest importance were excluded and a weighted decision tree analysis was applied to obtain an interpretable classifier. Area under the ROC curve (AUC), accuracy and generalization ability of the model were estimated using cross validation techniques.

RESULTS

Average and 7 clock-hour RNFL thicknesses were the parameters with the highest importance. Correlation between parameter values and impact on classification displayed a stepped pattern for average thickness. Decision tree model revealed that average thickness lower than 82 µm was a high predictor for early glaucoma. Model scores had AUC of 0.953 (95%CI: 0.903-0998), with an accuracy of 89%.

CONCLUSION

Gradient boosting methods provide accurate and highly interpretable classifiers to discriminate between early glaucoma and healthy eyes. Average and 7-hour RNFL thicknesses have the best discriminant power.

Optimizing Survival Analysis of XGBoost for Ties to Predict Disease Progression of Breast Cancer

OBJECTIVE

Some excellent prognostic models based on survival analysis methods for breast cancer have been proposed and extensively validated, which provide an essential means for clinical diagnosis and treatment to improve patient survival. To analyze clinical and follow-up data of 12119 breast cancer patients, derived from the Clinical Research Center for Breast (CRCB) in West China Hospital of Sichuan University, we developed a gradient boosting algorithm, called EXSA, by optimizing survival analysis of XGBoost framework for ties to predict the disease progression of breast cancer.

METHODS

EXSA is based on the XGBoost framework in machine learning and the Cox proportional hazards model in survival analysis. By taking Efron approximation of partial likelihood function as a learning objective for ties, EXSA derives gradient formulas of a more precise approximation. It optimizes and enhances the ability of XGBoost for survival data with ties. After retaining 4575 patients (3202 cases for training, 1373 cases for test), we exploit the developed EXSA method to build an excellent prognostic model to estimate disease progress. Risk score of disease progress is evaluated by the model, and the risk grouping and continuous functions between risk scores and disease progress rate at 5- and 10-year are also demonstrated.

RESULTS

Experimental results on test set show that the EXSA method achieves competitive performance with concordance index of 0.83454, 5-year and 10-year AUC of 0.83851 and 0.78155, respectively.

CONCLUSION

The proposed EXSA method can be utilized as an effective method for survival analysis.

SIGNIFICANCE

The proposed method in this paper can provide an important means for follow-up data of breast cancer or other disease research.

Gradient-Boosting Classifiers Combining Vessel Density and Tissue Thickness Measurements for Classifying Early to Moderate Glaucoma

PURPOSE

To compare gradient-boosting classifier (GBC) analysis of optical coherence tomography angiography (OCTA)-measured vessel density (VD) and OCT-measured tissue thickness to standard OCTA VD and OCT thickness parameters for classifying healthy eyes and eyes with early to moderate glaucoma.

DESIGN

Comparison of diagnostic tools.

METHODS

A total of 180 healthy eyes and 193 glaucomatous eyes with OCTA and OCT imaging of the macula and optic nerve head (ONH) were studied. Four GBCs were evaluated that combined 1) all macula VD and thickness measurements (Macula GBC), 2) all ONH VD and thickness measurements (ONH GBC), 3) all VD measurements from the macula and ONH (vessel density GBC), and 4) all thickness measurements from the macula and ONH (thickness GBC). ROC curve (AUROC) analyses compared the diagnostic accuracy of GBCs to that of standard instrument-provided parameters. A fifth GBC that combined all parameters (full GBC) also was investigated.

RESULTS

GBCs had better diagnostic accuracy than standard OCTA and OCT parameters with AUROCs ranging from 0.90 to 0.93 and 0.64 to 0.91, respectively. The full GBC (AUROC = 0.93) performed significantly better than the ONH GBC (AUROC = 0.91; P = .036) and the vessel density GBC (AUROC = 0.90; P = .010). All other GBCs performed similarly. The mean relative influence of each parameter included in the full GBC identified a combination of macular thickness and ONH VD measurements as the greatest contributors.

CONCLUSIONS

GBCs that combine OCTA and OCT macula and ONH measurements can improve diagnostic accuracy for glaucoma detection compared to most but not all instrument provided parameters.

Machine learning models based on the dimensionality reduction of standard automated perimetry data for glaucoma diagnosis

INTRODUCTION

Visual field testing via standard automated perimetry (SAP) is a commonly used glaucoma diagnosis method. Applying machine learning techniques to the visual field test results, a valid clinical diagnosis of glaucoma solely based on the SAP data is provided. In order to reflect structural-functional patterns of glaucoma on the automated diagnostic models, we propose composite variables derived from anatomically grouped visual field clusters to improve the prediction performance. A set of machine learning-based diagnostic models are designed that implement different input data manipulation, dimensionality reduction, and classification methods.

METHODS

Visual field testing data of 375 healthy and 257 glaucomatous eyes were used to build the diagnostic models. Three kinds of composite variables derived from the Garway-Heath map and the glaucoma hemifield test (GHT) sector map were included in the input variables in addition to the 52 SAP visual filed locations. Dimensionality reduction was conducted to select important variables so as to alleviate high-dimensionality problems. To validate the proposed methods, we applied four classifiers-linear discriminant analysis, naïve Bayes classifier, support vector machines, and artificial neural networks-and four dimensionality reduction methods-Pearson correlation coefficient-based variable selection, Markov blanket variable selection, the minimum redundancy maximum relevance algorithm, and principal component analysis- and compared their classification performances.

RESULTS

For all tested combinations, the classification performance improved when the proposed composite variables and dimensionality reduction techniques were implemented. The combination of total deviation values, the GHT sector map, support vector machines, and Markov blanket variable selection obtains the best performance: an area under the receiver operating characteristic curve (AUC) of 0.912.

CONCLUSION

A glaucoma diagnosis model giving an AUC of 0.912 was constructed by applying machine learning techniques to SAP data. The results show that dimensionality reduction not only reduces dimensions of the input space but also enhances the classification performance. The variable selection results show that the proposed composite variables from visual field clustering play a key role in the diagnosis model.

Determination of CERES TOA fluxes using Machine learning algorithms. Part I: Classification and retrieval of CERES cloudy and clear scenes

Continuous monitoring of the Earth radiation budget (ERB) is critical to our understanding of the Earth's climate and its variability with time. The Clouds and the Earth's Radiant Energy System (CERES) instrument is able to provide a long record of ERB for such scientific studies. This manuscript, which is first of a two-part paper, describes the new CERES algorithm for improving the clear/cloudy scene classification without the use of coincident cloud imager data. This new CERES algorithm is based on a subset of modern artificial intelligence (AI) paradigm called Machine Learning (ML) algorithms. This paper describes development and application of the ML algorithm known as Random Forests (RF) which is used to classify CERES broadband footprint measurements into clear and cloudy scenes. Results from the RF analysis carried using the CERES Single Scanner Footprint (SSF) data for the months of January and July are presented in the manuscript. The daytime RF misclassification rate (MCR) shows relatively large values (>30%) for snow, sea ice and bright desert surface types while lower values of (<10%) for forest surface type. MCR values observed for the nighttime data in general show relatively larger values for most of the surface types compared to the daytime MCR values. The modified MCR values show lower values (< 4%) for most surface types after thin cloud data is excluded from the analysis. Sensitivity analysis shows that the number of input variables and decision trees used in the RF analysis has substantial influence in determining the classification error.

Ensemble of a subset of kNN classifiers

Combining multiple classifiers, known as ensemble methods, can give substantial improvement in prediction performance of learning algorithms especially in the presence of non-informative features in the data sets. We propose an ensemble of subset of kNN classifiers, ESkNN, for classification task in two steps. Firstly, we choose classifiers based upon their individual performance using the out-of-sample accuracy. The selected classifiers are then combined sequentially starting from the best model and assessed for collective performance on a validation data set. We use bench mark data sets with their original and some added non-informative features for the evaluation of our method. The results are compared with usual kNN, bagged kNN, random kNN, multiple feature subset method, random forest and support vector machines. Our experimental comparisons on benchmark classification problems and simulated data sets reveal that the proposed ensemble gives better classification performance than the usual kNN and its ensembles, and performs comparable to random forest and support vector machines.

Integration of genomic, transcriptomic and proteomic data identifies two biologically distinct subtypes of invasive lobular breast cancer

Invasive lobular carcinoma (ILC) is the second most frequently occurring histological breast cancer subtype after invasive ductal carcinoma (IDC), accounting for around 10% of all breast cancers. The molecular processes that drive the development of ILC are still largely unknown. We have performed a comprehensive genomic, transcriptomic and proteomic analysis of a large ILC patient cohort and present here an integrated molecular portrait of ILC. Mutations in CDH1 and in the PI3K pathway are the most frequent molecular alterations in ILC. We identified two main subtypes of ILCs: (i) an immune related subtype with mRNA up-regulation of PD-L1, PD-1 and CTLA-4 and greater sensitivity to DNA-damaging agents in representative cell line models; (ii) a hormone related subtype, associated with Epithelial to Mesenchymal Transition (EMT), and gain of chromosomes 1q and 8q and loss of chromosome 11q. Using the somatic mutation rate and eIF4B protein level, we identified three groups with different clinical outcomes, including a group with extremely good prognosis. We provide a comprehensive overview of the molecular alterations driving ILC and have explored links with therapy response. This molecular characterization may help to tailor treatment of ILC through the application of specific targeted, chemo- and/or immune-therapies.

Glaucomatous patterns in Frequency Doubling Technology (FDT) perimetry data identified by unsupervised machine learning classifiers

Purpose

The variational Bayesian independent component analysis-mixture model (VIM), an unsupervised machine-learning classifier, was used to automatically separate Matrix Frequency Doubling Technology (FDT) perimetry data into clusters of healthy and glaucomatous eyes, and to identify axes representing statistically independent patterns of defect in the glaucoma clusters.

Methods

FDT measurements were obtained from 1,190 eyes with normal FDT results and 786 eyes with abnormal FDT results from the UCSD-based Diagnostic Innovations in Glaucoma Study (DIGS) and African Descent and Glaucoma Evaluation Study (ADAGES). For all eyes, VIM input was 52 threshold test points from the 24-2 test pattern, plus age.

Results

FDT mean deviation was −1.00 dB (S.D. = 2.80 dB) and −5.57 dB (S.D. = 5.09 dB) in FDT-normal eyes and FDT-abnormal eyes, respectively (p<0.001). VIM identified meaningful clusters of FDT data and positioned a set of statistically independent axes through the mean of each cluster. The optimal VIM model separated the FDT fields into 3 clusters. Cluster N contained primarily normal fields (1109/1190, specificity 93.1%) and clusters G₁ and G₂ combined, contained primarily abnormal fields (651/786, sensitivity 82.8%). For clusters G₁ and G₂ the optimal number of axes were 2 and 5, respectively. Patterns automatically generated along axes within the glaucoma clusters were similar to those known to be indicative of glaucoma. Fields located farther from the normal mean on each glaucoma axis showed increasing field defect severity.

Conclusions

VIM successfully separated FDT fields from healthy and glaucoma eyes without a priori information about class membership, and identified familiar glaucomatous patterns of loss.

Ordinal response prediction using bootstrap aggregation, with application to a high-throughput methylation data set

Many investigators conducting translational research are performing high-throughput genomic experiments and then developing multigenic classifiers using the resulting high-dimensional data set. In a large number of applications, the class to be predicted may be inherently ordinal. Examples of ordinal outcomes include tumor-node-metastasis (TNM) stage (I, II, III, IV); drug toxicity evaluated as none, mild, moderate, or severe; and response to treatment classified as complete response, partial response, stable disease, or progressive disease. While one can apply nominal response classification methods to ordinal response data, in doing so some information is lost that may improve the predictive performance of the classifier. This study examined the effectiveness of alternative ordinal splitting functions combined with bootstrap aggregation for classifying an ordinal response. We demonstrate that the ordinal impurity and ordered twoing methods have desirable properties for classifying ordinal response data and both perform well in comparison to other previously described methods. Developing a multigenic classifier is a common goal for microarray studies, and therefore application of the ordinal ensemble methods is demonstrated on a high-throughput methylation data set.

Angiopoietin-2 levels are associated with disease progression in metastatic malignant melanoma

PURPOSE

The blood vessel-destabilizing Tie2 ligand angiopoietin-2 (Ang-2) acts in concert with the vascular endothelial growth factor/vascular endothelial growth factor receptor system to control vessel assembly during tumor progression. We hypothesized that circulating soluble Ang-2 (sAng-2) may be involved in melanoma progression.

EXPERIMENTAL DESIGN

Serum samples (n=98) from melanoma patients (American Joint Committee on Cancer stages I-IV), biopsies of corresponding patients, and human melanoma cell lines were analyzed for expression of Ang-2 and S100beta. Multiple sera of a subcohort of 33 patients were tested during progression from stage III to IV. Small interfering RNA-based loss-of-function experiments were done to assess effects of Ang-2 on melanoma cells.

RESULTS

Circulating levels of sAng-2 correlate with tumor progression in melanoma patients (P<0.0001) and patient survival (P=0.007). Analysis of serum samples during the transition from stage III to IV identified an increase of sAng-2 up to 400%. Comparative analyses revealed a 56% superiority of sAng-2 as predictive marker over the established marker S100beta. Immunohistochemistry and reverse transcription-PCR confirmed the prominent expression of Ang-2 by tumor-associated endothelial cells but identified Ang-2 also as a secreted product of melanoma cells themselves. Corresponding cellular experiments revealed that human melanoma-isolated tumor cells were Tie2 positive and that Ang-2 acted as an autocrine regulator of melanoma cell migration and invasion.

CONCLUSIONS

The experiments establish sAng-2 as a biomarker of melanoma progression and metastasis correlating with tumor load and overall survival. The identification of an autocrine angiopoietin/Tie loop controlling melanoma migration and invasion warrants further functional experiments and validate the angiopoietin/Tie system as a promising therapeutic target for human melanomas.

Predicting progressive glaucomatous optic neuropathy using baseline standard automated perimetry data

PURPOSE

To test the hypothesis that specific locations and patterns of threshold findings within the visual field have predictive value for progressive glaucomatous optic neuropathy (pGON).

METHODS

Age-adjusted standard automated perimetry thresholds, along with other clinical variables gathered at the initial examination of 168 individuals with high-risk ocular hypertension or early glaucoma, were used as predictors in a classification tree model. The classification variable was a determination of pGON, based on longitudinally gathered stereo optic nerve head photographs. Only data for the worse eye of each individual were included. Data from 100 normal subjects were used to test the specificity of the models.

RESULTS

Classification tree models suggest that patterns of baseline visual field findings are predictive of pGON with sensitivity 65% and specificity 87% on average. Average specificity when data from normal subjects were run on the models was 69%.

CONCLUSIONS

Classification trees can be used to determine which visual field locations are most predictive of poorer prognosis for pGON. Spatial patterns within the visual field convey useable predictive information, in most cases when thresholds are still well within the classically defined normal range.

Glaucoma Detection With Frequency Doubling Perimetry and Short-wavelength Perimetry

PURPOSE

The aim of this analysis was to evaluate the diagnostic usefulness of frequency doubling technology (FDT) perimetry and short-wavelength perimetry (SWAP). Moreover, to study a combination of both methods using the machine-learning technique double-bagging, which was recently established in glaucoma research.

METHODS

Forty-three patients with "preperimetric" open-angle glaucoma (glaucomatous optic disc atrophy and no visual field defect in standard perimetry), 26 patients with "perimetric" open angle glaucoma (glaucomatous optic disc atrophy and visual field defect in standard perimetry), and 40 control subjects had FDT screening (protocol: C-20-5) and SWAP (Octopus 101, G2). Criteria for exclusion were color vision abnormalities, media opacities, and an age below 31 years or above 63 years. Data of 1 eye of each patient and control subject entered the statistical evaluation. A point wise evaluation of the diagnostic power of SWAP values was performed to derive spatial patterns of visual field loss. A double-bagging machine-learning algorithm was used to train classification rules on the basis of a combination of FDT scores and nerve fiber related visual field losses in SWAP. The diagnostic power of the classifiers was compared regarding their misclassification error rates and area under the receiver-operating characteristic curve.

RESULTS

The combination of FDT perimetry and SWAP yielded better diagnostic results compared with FDT or SWAP separately. The overall estimated misclassification error rate of the combined classifier was 24% compared with 28% for both SWAP and FDT perimetry. Regarding the estimated performance of classifier at high specificities (>80%) in control eyes as measured by the partial area under the receiver-operating characteristic curve, the combination of both instruments is also superior to the individual instruments.

CONCLUSIONS

A combination of SWAP and FDT perimetry, each targeting different neuronal pathways, may improve early glaucoma detection.

Comparing regularized B-spline neural network, multilayer perceptron and boosted-CART on two problems of heart arrhythmia diagnosis

Medical diagnosis has special requirements on reliability and interpretability of a learning scheme. Two problems in heart arrhythmia diagnosis are discussed in the paper 1) distinguish premature ventricular contraction beats from normal beats (Problem A); 2) distinguish premature ventricular contraction beats from premature atrial beats (Problem B). Analysis of the real clinical data shows that these two problems have different noise levels, which is suitable for addressing various requirements of medical domain. The performances and characteristics of three methods are compared: Regularized B-spline Neural Networks, Multilayer Perceptron and Boosted-CART using AdaBoost. Regularized B-spline Neural Network outperforms Multilayer Perceptron on the difficult Problem B, which suggests its potential in modeling complex system. Overall, Boosted-CART achieves best recognition rate with intermediate interpretability.

A spatio-temporal Bayesian network classifier for understanding visual field deterioration

OBJECTIVE

Progressive loss of the field of vision is characteristic of a number of eye diseases such as glaucoma which is a leading cause of irreversible blindness in the world. Recently, there has been an explosion in the amount of data being stored on patients who suffer from visual deterioration including field test data, retinal image data and patient demographic data. However, there has been relatively little work in modelling the spatial and temporal relationships common to such data. In this paper we introduce a novel method for classifying visual field (VF) data that explicitly models these spatial and temporal relationships.

METHODOLOGY

We carry out an analysis of our proposed spatio-temporal Bayesian classifier and compare it to a number of classifiers from the machine learning and statistical communities. These are all tested on two datasets of VF and clinical data. We investigate the receiver operating characteristics curves, the resulting network structures and also make use of existing anatomical knowledge of the eye in order to validate the discovered models.

RESULTS

Results are very encouraging showing that our classifiers are comparable to existing statistical models whilst also facilitating the understanding of underlying spatial and temporal relationships within VF data. The results reveal the potential of using such models for knowledge discovery within ophthalmic databases, such as networks reflecting the 'nasal step', an early indicator of the onset of glaucoma.

CONCLUSION

The results outlined in this paper pave the way for a substantial program of study involving many other spatial and temporal datasets, including retinal image and clinical data.

Automated segmentation of the optic nerve head for diagnosis of glaucoma

Glaucoma is the second most common cause of blindness worldwide. Low awareness and high costs connected to glaucoma are reasons to improve methods of screening and therapy. A well-established method for diagnosis of glaucoma is the examination of the optic nerve head using scanning-laser-tomography. This system acquires and analyzes the surface topography of the optic nerve head. The analysis that leads to a diagnosis of the disease depends on prior manual outlining of the optic nerve head by an experienced ophthalmologist. Our contribution presents a method for optic nerve head segmentation and its validation. The method is based on morphological operations, Hough transform, and an anchored active contour model. The results were validated by comparing the performance of different classifiers on data from a case-control study with contours of the optic nerve head manually outlined by an experienced ophthalmologist. We achieved the following results with respect to glaucoma diagnosis: linear discriminant analysis with 27.7% estimated error rate for automated segmentation (aut) and 26.8% estimated error rate for manual segmentation (man), classification trees with 25.2% (aut) and 22.0% (man) and bootstrap aggregation with 22.2% (aut) and 13.4% (man). It could thus be shown that our approach is suitable for automated diagnosis and screening of glaucoma.

Relevance vector machine and support vector machine classifier analysis of scanning laser polarimetry retinal nerve fiber layer measurements

PURPOSE

To classify healthy and glaucomatous eyes using relevance vector machine (RVM) and support vector machine (SVM) learning classifiers trained on retinal nerve fiber layer (RNFL) thickness measurements obtained by scanning laser polarimetry (SLP).

METHODS

Seventy-two eyes of 72 healthy control subjects (average age = 64.3 +/- 8.8 years, visual field mean deviation = -0.71 +/- 1.2 dB) and 92 eyes of 92 patients with glaucoma (average age = 66.9 +/- 8.9 years, visual field mean deviation = -5.32 +/- 4.0 dB) were imaged with SLP with variable corneal compensation (GDx VCC; Laser Diagnostic Technologies, San Diego, CA). RVM and SVM learning classifiers were trained and tested on SLP-determined RNFL thickness measurements from 14 standard parameters and 64 sectors (approximately 5.6 degrees each) obtained in the circumpapillary area under the instrument-defined measurement ellipse (total 78 parameters). Ten-fold cross-validation was used to train and test RVM and SVM classifiers on unique subsets of the full 164-eye data set and areas under the receiver operating characteristic (AUROC) curve for the classification of eyes in the test set were generated. AUROC curve results from RVM and SVM were compared to those for 14 SLP software-generated global and regional RNFL thickness parameters. Also reported was the AUROC curve for the GDx VCC software-generated nerve fiber indicator (NFI).

RESULTS

The AUROC curves for RVM and SVM were 0.90 and 0.91, respectively, and increased to 0.93 and 0.94 when the training sets were optimized with sequential forward and backward selection (resulting in reduced dimensional data sets). AUROC curves for optimized RVM and SVM were significantly larger than those for all individual SLP parameters. The AUROC curve for the NFI was 0.87.

CONCLUSIONS

Results from RVM and SVM trained on SLP RNFL thickness measurements are similar and provide accurate classification of glaucomatous and healthy eyes. RVM may be preferable to SVM, because it provides a Bayesian-derived probability of glaucoma as an output. These results suggest that these machine learning classifiers show good potential for glaucoma diagnosis.

Classification and regression tree analysis in public health: methodological review and comparison with logistic regression

BACKGROUND

Audience segmentation strategies are of increasing interest to public health professionals who wish to identify easily defined, mutually exclusive population subgroups whose members share similar characteristics that help determine participation in a health-related behavior as a basis for targeted interventions. Classification and regression tree (C&RT) analysis is a nonparametric decision tree methodology that has the ability to efficiently segment populations into meaningful subgroups. However, it is not commonly used in public health.

PURPOSE

This study provides a methodological overview of C&RT analysis for persons unfamiliar with the procedure.

METHODS AND RESULTS

An example of a C&RT analysis is provided and interpretation of results is discussed. Results are validated with those obtained from a logistic regression model that was created to replicate the C&RT findings. Results obtained from the example C&RT analysis are also compared to those obtained from a common approach to logistic regression, the stepwise selection procedure. Issues to consider when deciding whether to use C&RT are discussed, and situations in which C&RT may and may not be beneficial are described.

CONCLUSIONS

C&RT is a promising research tool for the identification of at-risk populations in public health research and outreach.

Automated detection of immunofluorescently labeled cytomegalovirus-infected cells in isolated peripheral blood leukocytes using decision tree analysis

BACKGROUND

Cytomegalovirus (CMV) infection continues to be a major problem for immunocompromised patients. Detection of viral antigens in leukocytes (antigenemia assay) is widely used for the diagnosis of CMV infection and for guiding antiviral therapy. The antigenemia technique, contingent upon the manual microscopic analysis of rare cells, is a laborious task that is subject to human error. In this study, we combine automated microscopy with artificial intelligence for reliable detection of fluorescently labeled CMV-infected cells.

METHODS

Cytospin preparations of peripheral blood leukocytes were immunofluorescently labeled for the CMV lower matrix phosphoprotein (pp65) and scanned in the Rare Event Imaging System (REIS), a fully automated image cytometer. The REIS detected potential positive objects and digitally recorded 49 measured cellular features for each identified case. The measurement data of these objects were analyzed by the See5 decision tree (DT) algorithm to ascertain whether they were true-positive detections.

RESULTS

The DT was built from the measurement data of 2,047 true- and 2,028 false-positive detections, collected from 32 patient samples. By designating misclassifications of false-negatives three times more costly, the 10-fold cross-validation sensitivity, specificity, and misclassification error of the assay was 94.3%, 56.2%, and 25%, respectively. The method was also validated using an independent test set of 21 patient samples, in which similar results were obtained.

CONCLUSIONS

To our knowledge, this study represents the first attempt to improve the accuracy of rare event image cytometry through the implementation of artificial intelligence methodology. Results suggest that cost-sensitive decision tree analysis of digitally measured cellular features vastly improves the performance of rare event image cytometry.