Two-stage Classification Strategy Research Articles

Two-stage classification strategy for breast cancer diagnosis using ultrasound-guided diffuse optical tomography and deep learning.

Ultrasound (US)-guided diffuse optical tomography (DOT) has demonstrated great potential for breast cancer diagnosis in which real-time or near real-time diagnosis with high accuracy is desired. We aim to use US-guided DOT to achieve an automated, fast, and accurate classification of breast lesions. We propose a two-stage classification strategy with deep learning. In the first stage, US images and histograms created from DOT perturbation measurements are combined to predict benign lesions. Then the non-benign suspicious lesions are passed through to the second stage, which combine US image features, DOT histogram features, and 3D DOT reconstructed images for final diagnosis. The first stage alone identified 73.0% of benign cases without image reconstruction. In distinguishing between benign and malignant breast lesions in patient data, the two-stage classification approach achieved an area under the receiver operating characteristic curve of 0.946, outperforming the diagnoses of all single-modality models and of a single-stage classification model that combines all US images, DOT histogram, and imaging features. The proposed two-stage classification strategy achieves better classification accuracy than single-modality-only models and a single-stage classification model that combines all features. It can potentially distinguish breast cancers from benign lesions in near real-time.

Ultra-low complexity random forest for optical fiber communications.

In this paper, we present an efficient equalizer based on random forest for channel equalization in optical fiber communication systems. The results are experimentally demonstrated in a 120 Gb/s, 375 km, dual-polarization 64-quadrature magnitude modulation (QAM) optical fiber communication platform. Based on the optimal parameters, we choose a series of deep learning algorithms for comparison. We find that random forest has the same level of equalization performance as deep neural networks as well as lower computational complexity. Moreover, we propose a two-step classification mechanism. We first divide the constellation points into two regions and then use different random forest equalizers to compensate the points in different regions. Based on this strategy, the system complexity and performance can be further reduced and improved. Furthermore, due to the plurality voting mechanism and two-stage classification strategy, the random forest-based equalizer can be applied to actual optical fiber communication systems.

Density Peaks Clustering Based on Feature Reduction and Quasi-Monte Carlo

Density peaks clustering (DPC) is a well-known density-based clustering algorithm that can deal with nonspherical clusters well. However, DPC has high computational complexity and space complexity in calculating local density ρ and distance δ , which makes it suitable only for small-scale data sets. In addition, for clustering high-dimensional data, the performance of DPC still needs to be improved. High-dimensional data not only make the data distribution more complex but also lead to more computational overheads. To address the above issues, we propose an improved density peaks clustering algorithm, which combines feature reduction and data sampling strategy. Specifically, features of the high-dimensional data are automatically extracted by principal component analysis (PCA), auto-encoder (AE), and t-distributed stochastic neighbor embedding (t-SNE). Next, in order to reduce the computational overhead, we propose a novel data sampling method for the low-dimensional feature data. Firstly, the data distribution in the low-dimensional feature space is estimated by the Quasi-Monte Carlo (QMC) sequence with low-discrepancy characteristics. Then, the representative QMC points are selected according to their cell densities. Next, the selected QMC points are used to calculate ρ and δ instead of the original data points. In general, the number of the selected QMC points is much smaller than that of the initial data set. Finally, a two-stage classification strategy based on the QMC points clustering results is proposed to classify the original data set. Compared with current works, our proposed algorithm can reduce the computational complexity from O n 2 to O N n , where N denotes the number of selected QMC points and n is the size of original data set, typically N ≪ n . Experimental results demonstrate that the proposed algorithm can effectively reduce the computational overhead and improve the model performance.

A two-stage classification model integrating feature fusion for coronary artery disease detection and classification

According to the World Health Organization, Coronary Artery Disease (CAD) is a leading cause of death globally. CAD is categorized into three types, namely Single Vessel Coronary Artery Disease (SVCAD), Double Vessel Coronary Artery Disease (DVCAD), and Triple Vessel Coronary Artery Disease (TVCAD). At present, angiography is the most popular technique to detect CAD that is quite expensive and invasive. Phonocardiogram (PCG), being economical and non-invasive, is a crucial modality towards the detection of cardiac disorders, but only trained medical professionals can interpret heart auscultations in clinical environments. This research aims to detect CAD and its types from PCG signatures through feature fusion and a two-stage classification strategy. The self-developed low-cost stethoscope was used to collect PCG data from a local hospital. The PCG signals were preprocessed through an iterative signal decomposition method known as Empirical Mode Decomposition (EMD). EMD decomposes the raw PCG signal into its constituent components called Intrinsic Mode Functions (IMFs). Preprocessed PCG signal was generated exclusively through combining those signal components that contain high discriminative characteristics and less redundancy. Next, Mel Frequency Cepstral Coefficients (MFCCs), spectral and statistical features were extracted. A two-stage classification framework was devised to identify healthy and CAD types. The first stage framework relies on the fusion of MFCC and statistical features with the K-nearest neighbor classifier to predict normal and CAD cases. The second stage is activated only when the first stage detects CAD. The fusion of spectral, statistical, and MFCC features was employed with Support Vector Machines classifier to categorize PCG signatures into DVCAD, SVCAD, and TVCAD classes in the second stage. The proposed method yields mean accuracy values of 88.0%, 89.2%, 91.1%, and 85.3% for normal, DVCAD, SVCAD, and TVCAD, respectively, through 10-fold cross-validation. Comparative analysis with existing approaches confirmed the reliability of the proposed method for categorizing CAD in general clinical environments. The proposed model enhances the diagnosis performance by providing a second opinion during the medical examination.

Auxiliary criterion conversion via spatiotemporal semantic encoding and feature entropy for action recognition

Video-based action recognition in realistic scenes is a core technology for human–computer interaction and smart surveillance. Although the trajectory features with the bag of visual words have confirmed promising performance, spatiotemporal interactive information cannot be effectively encoded which is valuable for classification. To address this issue, we propose a spatiotemporal semantic feature (ST-SF) and implement the conversion of it to the auxiliary criterion based on the information entropy theory. First, we present a text-based relevance analysis method to estimate the textual labels of objects most relevant to actions, which are employed to train the more targeted detectors based on the deep network. False detections are optimized by the inter-frame cooperativity and dynamic programming to construct the valid tubes. Then, we design the ST-SF to encode the interactive information, and the concept and calculation of feature entropy are defined based on the spatial distribution of ST-SFs on the training set. Finally, we achieve a two-stage classification strategy using the resulting decision gains. Experimental results on three publicly available datasets demonstrate that our method is robust and improves upon the state-of-the-art algorithms.

Robust detection of event-related potentials in a user-voluntary short-term imagery task.

Event-related potentials (ERPs) represent neuronal activity in the brain elicited by external visual or auditory stimulation and are widely used in brain-computer interface (BCI) systems. The ERP responses are elicited a few milliseconds after attending to an oddball stimulus; target and non-target stimuli are repeatedly flashed, and the ERP trials are averaged over time in order to improve their decoding accuracy. To reduce this time-consuming process, previous studies have attempted to evoke stronger ERP responses by changing certain experimental parameters like color, size, or the use of a face image as a target symbol. Since these exogenous potentials can be naturally evoked by merely looking at a target symbol, the BCI system could generate unintended commands while subjects are gazing at one of the symbols in a non-intentional mental state. We approached this problem of unintended command generation by assuming that a greater effort by the user in a short-term imagery task would evoke a discriminative ERP response. Three tasks were defined: passive attention, counting, and pitch-imagery. Users were instructed to passively attend to a target symbol, or to perform a mental tally of the number of target presentations, or to perform the novel task of imagining a high-pitch tone when the target symbol was highlighted. The decoding accuracy were 71.4%, 83.5%, and 89.2% for passive attention, counting, and pitch-imagery, respectively, after the fourth averaging procedure. We found stronger deflections in the N500 component corresponding to the levels of mental effort (passive attention: -1.094 ±0.88 μV, counting: -2.226 ±0.97 μV, and pitch-imagery: -2.883 ±0.74 μV), which highly influenced the decoding accuracy. In addition, the rate of binary classification between passive attention and pitch-imagery tasks was 73.5%, which is an adequate classification rate that motivated us to propose a two-stage classification strategy wherein the target symbols are estimated in the first stage and the passive or active mental state is decoded in the second stage. In this study, we found that the ERP response and the decoding accuracy are highly influenced by the user’s voluntary mental tasks. This could lead to a useful approach in practical ERP systems in two respects. Firstly, the user-voluntary tasks can be easily utilized in many different types of BCI systems, and performance enhancement is less dependent on the manipulation of the system’s external, visual stimulus parameters. Secondly, we propose an ERP system that classifies the brain state as intended or unintended by considering the measurable differences between passively gazing and actively performing the pitch-imagery tasks in the EEG signal thus minimizing unintended commands to the BCI system.

Complete three‐phase detection framework for identifying abnormal cervical cells

Automatic identification of abnormal cervical cells, including feature representation, feature combination and classification strategy, is highly demanded in women's annual cervical cancer screenings. However, previous methods only deal with one or two of these three phases, and currently there is few complete framework for this problem. A novel three-phrase boosting framework is proposed for the detection of abnormal cells from cervical smear images. First, the authors extract 160 dimensional features with respect to each cervical cell from three aspects, including cytology morphology, chromatin pathology and region intensity. In particular, 106 dimensional chromatin pathology features are newly adopted to describe the nucleus textural transformation. Second, an adaptive feature combination method is introduced to select the optimal feature patterns, which can combine all features using a reinforced margin-based approach with the heuristic knowledge. Finally, a two-stage classification strategy is presented to reduce erroneous classification abnormal cells using two different classifiers. Experimental results achieve state-of-the-art performance and the proposed framework outperforms the other 16 compared detection methods.

Feature selection using a one dimensional naïve Bayes' classifier increases the accuracy of support vector machine classification of CDR3 repertoires.

MotivationSomatic DNA recombination, the hallmark of vertebrate adaptive immunity, has the potential to generate a vast diversity of antigen receptor sequences. How this diversity captures antigen specificity remains incompletely understood. In this study we use high throughput sequencing to compare the global changes in T cell receptor β chain complementarity determining region 3 (CDR3β) sequences following immunization with ovalbumin administered with complete Freund’s adjuvant (CFA) or CFA alone.ResultsThe CDR3β sequences were deconstructed into short stretches of overlapping contiguous amino acids. The motifs were ranked according to a one-dimensional Bayesian classifier score comparing their frequency in the repertoires of the two immunization classes. The top ranking motifs were selected and used to create feature vectors which were used to train a support vector machine. The support vector machine achieved high classification scores in a leave-one-out validation test reaching >90% in some cases.SummaryThe study describes a novel two-stage classification strategy combining a one-dimensional Bayesian classifier with a support vector machine. Using this approach we demonstrate that the frequency of a small number of linear motifs three amino acids in length can accurately identify a CD4 T cell response to ovalbumin against a background response to the complex mixture of antigens which characterize Complete Freund’s Adjuvant.Availability and implementationThe sequence data is available at www.ncbi.nlm.nih.gov/sra/?term¼SRP075893. The Decombinator package is available at github.com/innate2adaptive/Decombinator. The R package e1071 is available at the CRAN repository https://cran.r-project.org/web/packages/e1071/index.html.Supplementary information Supplementary data are available at Bioinformatics online.

Vehicle Type Recognition Combining Global and Local Features via Two-Stage Classification

This study proposes a new vehicle type recognition method that combines global and local features via a two-stage classification. To extract the continuous and complete global feature, an improved Canny edge detection algorithm with smooth filtering and non-maxima suppression abilities is proposed. To extract the local feature from four partitioned key patches, a set of Gabor wavelet kernels with five scales and eight orientations is introduced. Different from the single-stage classification, where all features are incorporated into one classifier simultaneously, the proposed two-stage classification strategy leverages two types of features and classifiers. In the first stage, the preliminary recognition of large vehicle or small vehicle is conducted based on the global feature via a k-nearest neighbor probability classifier. Based on the preliminary result, the specific recognition of bus, truck, van, or sedan is achieved based on the local feature via a discriminative sparse representation based classifier. We experiment with the proposed method on the public and established datasets involving various challenging cases, such as partial occlusion, poor illumination, and scale variation. Experimental results show that the proposed method outperforms existing state-of-the-art methods.

Transmodal Learning of Functional Networks for Alzheimer's Disease Prediction.

Functional connectivity describes neural activity from resting-state functional magnetic resonance imaging (rs-fMRI). This noninvasive modality is a promising imaging biomarker of neurodegenerative diseases, such as Alzheimer's disease (AD), where the connectome can be an indicator to assess and to understand the pathology. However, it only provides noisy measurements of brain activity. As a consequence, it has shown fairly limited discrimination power on clinical groups. So far, the reference functional marker of AD is the fluorodeoxyglucose positron emission tomography (FDG-PET). It gives a reliable quantification of metabolic activity, but it is costly and invasive. Here, our goal is to analyze AD populations solely based on rs-fMRI, as functional connectivity is correlated to metabolism. We introduce transmodal learning: leveraging a prior from one modality to improve results of another modality on different subjects. A metabolic prior is learned from an independent FDG-PET dataset to improve functional connectivity-based prediction of AD. The prior acts as a regularization of connectivity learning and improves the estimation of discriminative patterns from distinct rs-fMRI datasets. Our approach is a two-stage classification strategy that combines several seed-based connectivity maps to cover a large number of functional networks that identify AD physiopathology. Experimental results show that our transmodal approach increases classification accuracy compared to pure rs-fMRI approaches, without resorting to additional invasive acquisitions. The method successfully recovers brain regions known to be impacted by the disease.

In search of the ideal risk-scoring system for very high-risk cardiac surgical patients: a two-stage approach.

BackgroundCardiac surgery patients at very high risk are difficult to stratify with the existing risk scores. The objective of this study is to assess the clinical performance of two existing risk stratification scores (EuroSCORE II and ACEF score) in the setting of very high-risk patients undergoing cardiac surgery, and to identify a possible strategy to better address this patient population.MethodsThree-thousand-four-hundred-twenty eight (3,428) consecutive adult patients receiving cardiac operations at a single institution were investigated. Patients having an operative mortality risk >25 % at either the EuroSCORE II or the ACEF score were selected (105 patients). The discrimination power and calibration of the EuroSCORE II and the ACEF score were investigated. Factors associated with operative mortality were included in a multivariable logistic regression model and a new model was re-built for this patient population.ResultsThe observed mortality rate was 26 %. The expected mortality rate was underestimated by the EuroSCORE II (16 %) and overestimated by the ACEF Score (36 %). The EuroSCORE maintained a good discrimination (c-statistics 0.75) while the ACEF score did not (c-statistics 0.52). Within this patient population, the independent risk factors for operative mortality were emergency surgery, serum creatinine levels, pulmonary hypertension, and preoperative anemia. A model based on these factors provided an expected mortality risk of 26 % with a good discrimination (c-statics 0.82). Applying this model to extremely high-risk patients (expected mortality rate > 50 %) resulted in the re-classification of 25 % of the patient population.ConclusionsThe existing risk models have a poor clinical relevance in the segment of patients at very high mortality risk. This is particularly frustrating, because these patients are those where the decision-making process is more important. A two-stage classification strategy (first stage: EuroSCORE II/ACEF score risk > 25 %; second stage: reclassification based on pulmonary hypertension, serum creatinine, and anemia) seems a possible strategy to correctly address very high-risk patients.

Bayesian nonparametric classification for spectroscopy data

High-dimensional spectroscopy data are increasingly common in many fields of science. Building classification models in this context is challenging, due not only to high dimensionality but also to high autocorrelations. A two-stage classification strategy is proposed. First, in a data pre-processing step, the dimensionality of the data is reduced using one of two distinct methods. The output of either of these methods is then used to feed a classification procedure that uses a multivariate density estimate from a Bayesian nonparametric mixture model for discrimination purposes. The model employed is based on a random probability measure with decreasing weights. This nonparametric prior is chosen so as to ease the identifiability and label switching problems inherent to these models. This simple and flexible classification strategy is applied to the well-known ‘meat’ data set. The results are similar or better than previously reported in the literature for the same data.

Semantic Scene Detection System for Baseball Videos

In this paper, we proposed a content-based multimedia analysis/retrieval system mainly which is capable of handling the high-level content analysis such as the semantic scene detection for baseball broadcast videos. Here, eight semantic scene classes were predefined for baseball videos. First, an effective shot boundary detection scheme based on scalable color histogram was proposed to segment a video into many shots. Then, various visual features including field color features, skin color features, and camera motion information were extracted to analyze the semantics for each shot. According to the visual properties of different scenes, we developed a two-stage classification strategy for the semantic scene detection. Finally, the experimental results showed that the proposed framework identifies eight semantic baseball scenes with precision 81% and recall 84%.

Comparing a Preliminary Racial Classification with a Numerical Classification of the Maize Landraces of Uruguay

Plant genetic diversity is a major component of any agricultural ecosystem. Thus, it is essential to classify genetic resources properly to conserve, evaluate, and enhance germplasm efficiently. In maize (Zea mays L.), many classification systems have been used for delineating maize races. From the 1980s, with the use of computers, numerical taxonomy became increasingly important and multivariate methods began to be used for classifying genetic resources. The objective of this study was to compare two methods of classification of Uruguayan maize landraces: (i) a preliminary racial classification obtained through visual assessment and (ii) a numerical classification. The numerical classification was conducted by means of a two-stage classification strategy: first, initial groups were formed by the Ward method and, next, the Modified Location Model (MLM) refined those groups. This classification was compared with the preliminary racial classification by four criteria. The Ward-MLM strategy generated more homogeneous groups than those corresponding to the preliminary racial classification. The numerical classification maintained the structure of the more differentiated races, but divided the Cateto Sulino race into two more homogeneous groups, each with smaller variance and more differentiated than other groups. Numerical classification produced groups with clearly distinct characteristics, in terms of the numerical variables, and better, in terms of the four criteria used, than those formed on the basis of racial classification. These results will be the basis for an improved racial classification of maize landraces of Uruguay.

Comparing a Preliminary Racial Classification with a Numerical Classification of the Maize Landraces of Uruguay

Plant genetic diversity is a major component of any agricultural ecosystem. Thus, it is essential to classify genetic resources properly to conserve, evaluate, and enhance germplasm efficiently. In maize (Zea mays L.), many classification systems have been used for delineating maize races. From the 1980s, with the use of computers, numerical taxonomy became increasingly important and multivariate methods began to be used for classifying genetic resources. The objective of this study was to compare two methods of classification of Uruguayan maize landraces: (i) a preliminary racial classification obtained through visual assessment and (ii) a numerical classification. The numerical classification was conducted by means of a two-stage classification strategy: first, initial groups were formed by the Ward method and, next, the Modified Location Model (MLM) refined those groups. This classification was compared with the preliminary racial classification by four criteria. The Ward-MLM strategy generated more homogeneous groups than those corresponding to the preliminary racial classification. The numerical classification maintained the structure of the more differentiated races, but divided the Cateto Sulino race into two more homogeneous groups, each with smaller variance and more differentiated than other groups. Numerical classification produced groups with clearly distinct characteristics, in terms of the numerical variables, and better, in terms of the four criteria used, than those formed on the basis of racial classification. These results will be the basis for an improved racial classification of maize landraces of Uruguay.