RESUMEN

Two people's physiological responses become more similar as those people talk or cooperate, a phenomenon called physiological synchrony. The degree of synchrony correlates with conversation engagement and cooperation quality, and could thus be used to characterize interpersonal interaction. In this study, we used a combination of physiological synchrony metrics and pattern recognition algorithms to automatically classify four different dyadic conversation scenarios: two-sided positive conversation, two-sided negative conversation, and two one-sided scenarios. Heart rate, skin conductance, respiration and peripheral skin temperature were measured from 16 dyads in all four scenarios, and individual as well as synchrony features were extracted from them. A two-stage classifier based on stepwise feature selection and linear discriminant analysis achieved a four-class classification accuracy of 75.0% in leave-dyad-out crossvalidation. Removing synchrony features reduced accuracy to 65.6%, indicating that synchrony is informative. In the future, such classification algorithms may be used to, e.g., provide real-time feedback about conversation mood to participants, with applications in areas such as mental health counseling and education. The approach may also generalize to group scenarios and adjacent areas such as cooperation and competition.

RESUMEN

BACKGROUND: In this study, we aimed to develop a new end-to-end learning model called Graph-Drug-Target Interaction (DTI), which integrates various types of information in the heterogeneous network data, and to explore automatic learning of the topology-maintaining representations of drugs and targets, thereby effectively contributing to the prediction of DTI. Precise predictions of DTI can guide drug discovery and development. Most machine learning algorithms integrate multiple data sources and combine them with common embedding methods. However, the relationship between the drugs and target proteins is not well reported. Although some existing studies have used heterogeneous network graphs for DTI prediction, there are many limitations in the neighborhood information between the nodes in the heterogeneous network graphs. We studied the drug-drug interaction (DDI) and DTI from DrugBank Version 3.0, protein-protein interaction (PPI) from the human protein reference database Release 9, drug structure similarity from Morgan fingerprints of radius 2 and calculated by RDKit, and protein sequence similarity from Smith-Waterman score. METHOD: Our study consists of three major components. First, various drugs and target proteins were integrated, and a heterogeneous network was established based on a series of data sets. Second, the graph neural networks-inspired graph auto-encoding method was used to extract high-order structural information from the heterogeneous networks, thereby revealing the description of nodes (drugs and proteins) and their topological neighbors. Finally, potential DTI prediction was made, and the obtained samples were sent to the classifier for secondary classification. RESULTS: The performance of Graph-DTI and all baseline methods was evaluated using the sums of the area under the precision-recall curve (AUPR) and the area under the receiver operating characteristic curve (AUC). The results indicated that Graph-DTI outperformed the baseline methods in both performance results. CONCLUSION: Compared with other baseline DTI prediction methods, the results showed that Graph-DTI had better prediction performance. Additionally, in this study, we effectively classified drugs corresponding to different targets and vice versa. The above findings showed that Graph-DTI provided a powerful tool for drug research, development, and repositioning. Graph-DTI can serve as a drug development and repositioning tool more effectively than previous studies that did not use heterogeneous network graph embedding.

RESUMEN

Transfer learning (TL) techniques can enable effective learning in data scarce domains by allowing one to re-purpose data or scientific knowledge available in relevant source domains for predictive tasks in a target domain of interest. In this Data in Brief article, we present a synthetic dataset for binary classification in the context of Bayesian transfer learning, which can be used for the design and evaluation of TL-based classifiers. For this purpose, we consider numerous combinations of classification settings, based on which we simulate a diverse set of feature-label distributions with varying learning complexity. For each set of model parameters, we provide a pair of target and source datasets that have been jointly sampled from the underlying feature-label distributions in the target and source domains, respectively. For both target and source domains, the data in a given class and domain are normally distributed, where the distributions across domains are related to each other through a joint prior. To ensure the consistency of the classification complexity across the provided datasets, we have controlled the Bayes error such that it is maintained within a range of predefined values that mimic realistic classification scenarios across different relatedness levels. The provided datasets may serve as useful resources for designing and benchmarking transfer learning schemes for binary classification as well as the estimation of classification error.

RESUMEN

Features extracted from the wavelet transform coefficient matrix are widely used in the design of machine learning models to classify event-related potential (ERP) and electroencephalography (EEG) signals in a wide range of brain activity research and clinical studies. This novel study is aimed at dramatically improving the performance of such wavelet-based classifiers by exploiting information offered by the cone of influence (COI) of the continuous wavelet transform (CWT). The COI is a boundary that is superimposed on the wavelet scalogram to delineate the coefficients that are accurate from those that are inaccurate due to edge effects. The features derived from the inaccurate coefficients are, therefore, unreliable. In this study, it is hypothesized that the classifier performance would improve if unreliable features, which are outside the COI, are zeroed out, and the performance would improve even further if those features are cropped out completely. The entire, zeroed out, and cropped scalograms are referred to as the "same" (S)-scalogram, "zeroed out" (Z)-scalogram, and the "valid" (V)-scalogram, respectively. The strategy to validate the hypotheses is to formulate three classification approaches in which the feature vectors are extracted from the (a) S-scalogram in the standard manner, (b) Z-scalogram, and (c) V-scalogram. A subsampling strategy is developed to generate small-sample ERP ensembles to enable customized classifier design for single subjects, and a strategy is developed to select a subset of channels from multiple ERP channels. The three scalogram approaches are implemented using support vector machines, random forests, k-nearest neighbor, multilayer perceptron neural networks, and deep learning convolution neural networks. In order to validate the performance hypotheses, experiments are designed to classify the multi-channel ERPs of five subjects engaged in distinguishing between synonymous and non-synonymous word pairs. The results confirm that the classifiers using the Z-scalogram features outperform those using the S-scalogram features, and the classifiers using the V-scalogram features outperform those using the Z-scalogram features. Most importantly, the relative improvement of the V-scalogram classifiers over the standard S-scalogram classifiers is dramatic. Additionally, enabling the design of customized classifiers for individual subjects is an important contribution to ERP/EEG-based studies and diagnoses of patient-specific disorders.

RESUMEN

The research on intelligent fault diagnosis has yielded remarkable achievements based on artificial intelligence-related technologies. In engineering scenarios, machines usually work in a normal condition, which means limited fault data can be collected. Intelligent fault diagnosis with small & imbalanced data (S&I-IFD), which refers to build intelligent diagnosis models using limited machine faulty samples to achieve accurate fault identification, has been attracting the attention of researchers. Nowadays, the research on S&I-IFD has achieved fruitful results, but a review of the latest achievements is still lacking, and the future research directions are not clear enough. To address this, we review the research results on S&I-IFD and provides some future perspectives in this paper. The existing research results are divided into three categories: the data augmentation-based, the feature learning-based, and the classifier design-based. Data augmentation-based strategy improves the performance of diagnosis models by augmenting training data. Feature learning-based strategy identifies faults accurately by extracting features from small & imbalanced data. Classifier design-based strategy achieves high diagnosis accuracy by constructing classifiers suitable for small & imbalanced data. Finally, this paper points out the research challenges faced by S&I-IFD and provides some directions that may bring breakthroughs, including meta-learning and zero-shot learning.

RESUMEN

In Machine Learning, feature selection is an important step in classifier design. It consists of finding a subset of features that is optimum for a given cost function. One possibility to solve feature selection is to organize all possible feature subsets into a Boolean lattice and to exploit the fact that the costs of chains in that lattice describe U-shaped curves. Minimization of such cost function is known as the U-curve problem. Recently, a study proposed U-Curve Search (UCS), an optimal algorithm for that problem, which was successfully used for feature selection. However, despite of the algorithm optimality, the UCS required time in computational assays was exponential on the number of features. Here, we report that such scalability issue arises due to the fact that the U-curve problem is NP-hard. In the sequence, we introduce the Parallel U-Curve Search (PUCS), a new algorithm for the U-curve problem. In PUCS, we present a novel way to partition the search space into smaller Boolean lattices, thus rendering the algorithm highly parallelizable. We also provide computational assays with both synthetic data and Machine Learning datasets, where the PUCS performance was assessed against UCS and other golden standard algorithms in feature selection.

RESUMEN

In Machine Learning, feature selection is an important step in classifier design. It consists of finding a subset of features that is optimum for a given cost function. One possibility to solve feature selection is to organize all possible feature subsets into a Boolean lattice and to exploit the fact that the costs of chains in that lattice describe U-shaped curves. Minimization of such cost function is known as the U-curve problem. Recently, a study proposed U-Curve Search (UCS), an optimal algorithm for that problem, which was successfully used for feature selection. However, despite of the algorithm optimality, the UCS required time in computational assays was exponential on the number of features. Here, we report that such scalability issue arises due to the fact that the U-curve problem is NP-hard. In the sequence, we introduce the Parallel U-Curve Search (PUCS), a new algorithm for the U-curve problem. In PUCS, we present a novel way to partition the search space into smaller Boolean lattices, thus rendering the algorithm highly parallelizable. We also provide computational assays with both synthetic data and Machine Learning datasets, where the PUCS performance was assessed against UCS and other golden standard algorithms in feature selection

RESUMEN

Most activity classifiers focus on recognizing application-specific activities that are mostly performed in a scripted manner, where there is very little room for variation within the activity. These classifiers are mainly good at recognizing short scripted activities that are performed in a specific way. In reality, especially when considering daily activities, humans perform complex activities in a variety of ways. In this work, we aim to make activity recognition more practical by proposing a novel approach to recognize complex heterogeneous activities that could be performed in a wide variety of ways. We collect data from 15 subjects performing eight complex activities and test our approach while analyzing it from different aspects. The results show the validity of our approach. They also show how it performs better than the state-of-the-art approaches that tried to recognize the same activities in a more controlled environment.

RESUMEN

The purpose of the present study was to employ a computer-aided diagnosis system that classifies chronic liver disease (CLD) using ultrasound shear wave elastography (SWE) imaging, with a stiffness value-clustering and machine-learning algorithm. A clinical data set of 126 patients (56 healthy controls, 70 with CLD) was analyzed. First, an RGB-to-stiffness inverse mapping technique was employed. A five-cluster segmentation was then performed associating corresponding different-color regions with certain stiffness value ranges acquired from the SWE manufacturer-provided color bar. Subsequently, 35 features (7 for each cluster), indicative of physical characteristics existing within the SWE image, were extracted. A stepwise regression analysis toward feature reduction was used to derive a reduced feature subset that was fed into the support vector machine classification algorithm to classify CLD from healthy cases. The highest accuracy in classification of healthy to CLD subject discrimination from the support vector machine model was 87.3% with sensitivity and specificity values of 93.5% and 81.2%, respectively. Receiver operating characteristic curve analysis gave an area under the curve value of 0.87 (confidence interval: 0.77-0.92). A machine-learning algorithm that quantifies color information in terms of stiffness values from SWE images and discriminates CLD from healthy cases is introduced. New objective parameters and criteria for CLD diagnosis employing SWE images provided by the present study can be considered an important step toward color-based interpretation, and could assist radiologists' diagnostic performance on a daily basis after being installed in a PC and employed retrospectively, immediately after the examination.

Asunto(s)

Diagnóstico por Computador/métodos , Diagnóstico por Imagen de Elasticidad/métodos , Hepatopatías/diagnóstico por imagen , Aprendizaje Automático , Adolescente , Anciano , Algoritmos , Enfermedad Crónica , Color , Femenino , Humanos , Hígado/diagnóstico por imagen , Masculino , Persona de Mediana Edad , Sensibilidad y Especificidad , Adulto Joven

RESUMEN

BACKGROUND: The time and frequency features of motor unit action potentials (MUAPs) extracted from electromyographic (EMG) signal provide discriminative information for diagnosis and treatment of neuromuscular disorders. However, the results of conventional automatic diagnosis methods using MUAP features is not convincing yet. OBJECTIVE: The main goal in designing a MUAP characterization system is obtaining high classification accuracy to be used in clinical decision system. For this aim, in this study, a robust classifier is proposed to improve MUAP classification performance in estimating the class label (myopathic, neuropathic and normal) of a given MUAP. METHOD: The proposed scheme employs both time and time-frequency features of a MUAP along with an ensemble of support vector machines (SVMs) classifiers in hybrid serial/parallel architecture. Time domain features includes phase, turn, peak to peak amplitude, area, and duration of the MUAP. Time-frequency features are discrete wavelet transform coefficients of the MUAP. RESULTS: Evaluation results of the developed system using EMG signals of 23 subjects (7 with myopathic, 8 with neuropathic and 8 with no diseases)  showed that the system estimated the class label of MUAPs extracted from these signals with average of accuracy of 91% which is at least 5% higher than the accuracy of two previously presented methods. CONCLUSION: Using different optimized subsets of features along with the presented hybrid classifier results in a classification accuracy that is encouraging to be used in clinical applications for MUAP characterization. 

RESUMEN

Hidden Markov Model (HMM) is a well established technique for detecting patterns in a stream of observations. It performs well when the observation sequence does not contain unseen patterns that were not part of the training set. In an unconstrained environment, the observation sequence might contain new patterns that the HMM model is not familiar with. In such cases, HMM will match the test pattern to some trained pattern, which is most similar to the test pattern. This increases the false positives in the system. In this paper, we are describing a threshold based technique to detect such irrelevant patterns in a continuous stream of observations, and classify them as unwanted or bad patterns. The novelty of our approach is that it allows early detection of unwanted patterns. Test patterns are validated on a fixed length substring of observation sequence, rather than on the whole observation sequence corresponding to the test pattern. The substrings are validated based on its similarity with a valid pattern using a threshold value. This reduces the latency of detection of unwanted movement, and makes the detection process independent of duration of the various pattern classes. We evaluated this technique in the context of body sensor networks based human action recognition, and have achieved about 93 percent accuracy in detecting unwanted actions, while maintaining a 94 percent accuracy of recognizing valid actions.