Multi-way Classification Research Articles

A Clinically Explainable AI-Based Grading System for Age-Related Macular Degeneration Using Optical Coherence Tomography.

We propose an automated, explainable artificial intelligence (xAI) system for age-related macular degeneration (AMD) diagnosis. Mimicking the physician's perceptions, the proposed xAI system is capable of deriving clinically meaningful features from optical coherence tomography (OCT) B-scan images to differentiate between a normal retina, different grades of AMD (early, intermediate, geographic atrophy (GA), inactive wet or active neovascular disease [exudative or wet AMD]), and non-AMD diseases. Particularly, we extract retinal OCT-based clinical imaging markers that are correlated with the progression of AMD, which include: (i) subretinal tissue, sub-retinal pigment epithelial tissue, intraretinal fluid, subretinal fluid, and choroidal hypertransmission detection using a DeepLabV3+ network; (ii) detection of merged retina layers using a novel convolutional neural network model; (iii) drusen detection based on 2D curvature analysis; (iv) estimation of retinal layers' thickness, and first-order and higher-order reflectivity features. Those clinical features are used to grade a retinal OCT in a hierarchical decision tree process. The first step looks for severe disruption of retinal layers' indicative of advanced AMD. These cases are analyzed further to diagnose GA, inactive wet AMD, active wet AMD, and non-AMD diseases. Less severe cases are analyzed using a different pipeline to identify OCT with AMD-specific pathology, which is graded as intermediate-stage or early-stage AMD. The remainder is classified as either being a normal retina or having other non-AMD pathology. The proposed system in the multi-way classification task, evaluated on 1285 OCT images, achieved 90.82% accuracy. These promising results demonstrated the capability to automatically distinguish between normal eyes and all AMD grades in addition to non-AMD diseases.

Further results on controlling the false discovery rate under some complex grouping structure of hypotheses

This article considers developing false discovery rate (FDR) controlling methods for testing multiple hypotheses under three different classification settings of the hypotheses into groups — simultaneous multi-way classification, hierarchical classification, and a combination of these two classifications. The methods are developed in their oracle forms by considering a weighted version of the Benjamini–Hochberg (BH, 1995) method, with the weights encoding the underlying structural information about the hypotheses. They control the FDR when the p-values involved are Positively Regression Dependent on the Subset (PRDS) of null p-values, and are more powerful than the BH procedure. Data-adaptive versions of these methods are also proposed by appropriately estimating the weights under the different types of classification. The proposed data-adaptive methods control the FDR at the desired level when the p-values are independent and, as simulations show, they can be more powerful FDR controlling methods, even under certain dependency, than some existing comparable multiple testing methods. We apply the data-adaptive method under the above-mentioned combined classification setting to analyze a publicly available neuro-imaging dataset. Such data typically have complex classification structures that were not addressed in previously available multiple testing methods, as far as we know. Our proposed method, with a flexible weighting scheme, is poised to utilize more information from the data in its decision making process, than other existing multiple testing methods.

A data mining approach to forecast students’ career placement probabilities and recommendations in the programming field

The career opportunities in computer programming are vast and rapidly increasing. Skilled software engineers, programmers, and developers are vigorously in demand worldwide. The capability to forecast a student's future career can be helpful in a wide variety of pedagogical practices. Data mining is becoming a more robust tool for analysis and forecasting. Therefore, to forecast career placement probabilities in the programming field, data mining classification and forecast techniques are used in this study to facilitate prospective students to make sensible career decisions. To achieve this objective, passed-out graduates' data is utilized, which comprises features like graduates' educational attainments in pre-university grades, i.e. grades of matriculation and intermediate, programming courses taught in early semesters along with the Cumulative Grade Point Average (CGPA) with the internship experience, gender, and family demographic information. Various multi-way Classification Trees are generated, which could help students to choose a branch with high career placement probabilities. From historical data, the Classification Trees have determined whether the branch is 'Good', 'Satisfactory', or 'Poor' based on the given information. The experimental findings indicate that all the features significantly influence the career placement probabilities in the programming field.

Multiway Sparse Distance Weighted Discrimination

Modern data often take the form of a multiway array. However, most classification methods are designed for vectors, that is, one-way arrays. Distance weighted discrimination (DWD) is a popular high-dimensional classification method that has been extended to the multiway context, with dramatic improvements in performance when data have multiway structure. However, the previous implementation of multiway DWD was restricted to classification of matrices, and did not account for sparsity. In this article, we develop a general framework for multiway classification which is applicable to any number of dimensions and any degree of sparsity. We conducted extensive simulation studies, showing that our model is robust to the degree of sparsity and improves classification accuracy when the data have multiway structure. For our motivating application, magnetic resonance spectroscopy (MRS) was used to measure the abundance of several metabolites across multiple neurological regions and across multiple time points in a mouse model of Friedreich’s ataxia, yielding a four-way data array. Our method reveals a robust and interpretable multi-region metabolomic signal that discriminates the groups of interest. We also successfully apply our method to gene expression time course data for multiple sclerosis treatment. An R implementation is available in the package MultiwayClassification at http://github.com/lockEF/MultiwayClassification. Supplementary materials for this article are available online.

Ultra-fine Entity Typing with Indirect Supervision from Natural Language Inference

Abstract The task of ultra-fine entity typing (UFET) seeks to predict diverse and free-form words or phrases that describe the appropriate types of entities mentioned in sentences. A key challenge for this task lies in the large number of types and the scarcity of annotated data per type. Existing systems formulate the task as a multi-way classification problem and train directly or distantly supervised classifiers. This causes two issues: (i) the classifiers do not capture the type semantics because types are often converted into indices; (ii) systems developed in this way are limited to predicting within a pre-defined type set, and often fall short of generalizing to types that are rarely seen or unseen in training. This work presents LITE🍻, a new approach that formulates entity typing as a natural language inference (NLI) problem, making use of (i) the indirect supervision from NLI to infer type information meaningfully represented as textual hypotheses and alleviate the data scarcity issue, as well as (ii) a learning-to-rank objective to avoid the pre-defining of a type set. Experiments show that, with limited training data, LITE obtains state-of-the-art performance on the UFET task. In addition, LITE demonstrates its strong generalizability by not only yielding best results on other fine-grained entity typing benchmarks, more importantly, a pre-trained LITE system works well on new data containing unseen types.1

Detection of adulterations in a valuable Brazilian honey by using spectrofluorimetry and multiway classification

Spectrofluorimetry combined with multiway chemometric tools were applied to discriminate pure Aroeira honey samples from samples adulterated with corn syrup, sugar cane molasses and polyfloral honey. Excitation emission spectra were acquired for 232 honey samples by recording excitation from 250 to 500 nm and emission from 270 to 640 nm. Parallel factor analysis (PARAFAC), partial least squares discriminant analysis (PLS-DA), unfolded PLS-DA (UPLS-DA) and multilinear PLS-DA (NPLS-DA) methods were used to decompose the spectral data and build classification models. PLS-DA models presented poor classification rates, demonstrating the limitation of the traditional two-way methods for this dataset, and leading to the development of three-way classification models. Overall, UPLS-DA provided the best classification results with misclassification rates of 4% and 8% for the training and test sets, respectively. These results showed the potential of the proposed method for routine laboratory analysis as a simple, reliable, and affordable tool.

Development and evaluation of an artificial intelligence system for COVID-19 diagnosis

Early detection of COVID-19 based on chest CT enables timely treatment of patients and helps control the spread of the disease. We proposed an artificial intelligence (AI) system for rapid COVID-19 detection and performed extensive statistical analysis of CTs of COVID-19 based on the AI system. We developed and evaluated our system on a large dataset with more than 10 thousand CT volumes from COVID-19, influenza-A/B, non-viral community acquired pneumonia (CAP) and non-pneumonia subjects. In such a difficult multi-class diagnosis task, our deep convolutional neural network-based system is able to achieve an area under the receiver operating characteristic curve (AUC) of 97.81% for multi-way classification on test cohort of 3,199 scans, AUC of 92.99% and 93.25% on two publicly available datasets, CC-CCII and MosMedData respectively. In a reader study involving five radiologists, the AI system outperforms all of radiologists in more challenging tasks at a speed of two orders of magnitude above them. Diagnosis performance of chest x-ray (CXR) is compared to that of CT. Detailed interpretation of deep network is also performed to relate system outputs with CT presentations. The code is available at https://github.com/ChenWWWeixiang/diagnosis_covid19.

Molecular fluorescence spectroscopy with multi-way analysis techniques detects spectral variations distinguishing uninfected serum versus dengue or chikungunya viral infected samples

Significant attempts are being made worldwide in an attempt to develop a tool that, with a simple analysis, is capable of distinguishing between different arboviruses. Herein, we employ molecular fluorescence spectroscopy as a sensitive and specific rapid tool, with simple methodology response, capable of identifying spectral variations between serum samples with or without the dengue or chikungunya viruses. Towards this, excitation emission matrices (EEM) of clinical samples from patients with dengue or chikungunya, in addition to uninfected controls, were separated into a training or test set and analysed using multi-way classification models such as n-PLSDA, PARAFAC-LDA and PARAFAC-QDA. Results were evaluated based on calculations of accuracy, sensitivity, specificity and F score; the most efficient model was identified to be PARAFAC-QDA, whereby 100% was obtained for all figures of merit. QDA was able to predict all samples in the test set based on the scores present in the factors selected by PARAFAC. The loadings obtained by PARAFAC can be used in future studies to prove the direct or indirect relationship of spectral changes caused by the presence of these viruses. This study demonstrates that molecular fluorescence spectroscopy has a greater capacity to detect spectral variations related to the presence of such viruses when compared to more conventional techniques.

Sentence Level Sentiment Analysis using Deep Learning Method

An Emoji is a small image representing facial expression, entity or a concept that can be either static or animated. In this paper, Emojis are used to study both cross-language and language based sentiment patterns. All the languages do not come with fair amount of labels. Emojis are useful signs of sentiment analysis in cross-lingual tweets. In this paper, an approach is proposed to extend the existing binary sentiment classification using multi-way classification. A novel Long Short-Term Memory (LSTM) - Convolutional Neural Networks (CNN) based model is proposed to obtain sentiments from emojis. The sentiments are classified using Deep Learning method like CNN. The proposed system outperforms the existing system in terms of Accuracy, Precision, Recall, F-Measure and Time Period. Finally the researcher manifests the fact that the CNN and LSTM combination as model shows an immense improvement to detecting the sentiment targets.

Developing an Excitation-Emission Matrix Fluorescence Spectroscopy Method Coupled with Multi-way Classification Algorithms for the Identification of the Adulteration of Shanxi Aged Vinegars

In this paper, an excitation-emission matrix (EEM) fluorescence spectroscopy method coupled with two kinds of multi-way classification algorithms was developed for the identification of the adulteration of Shanxi aged vinegars. The first classification model was built from selective information provided by parallel factor analysis (PARAFAC) based on partial least squares-discriminant analysis (PLS-DA) algorithm and the other one was modeled based on strictly multi-way classification algorithm multi-way partial least squares-discriminant analysis (N-PLS-DA). Both classification models showed good clustering tendency for authentic and adulterated Shanxi aged vinegars. Compared with the PARAFAC-PLS-DA method, more accurate and reliable classification results were provided by the N-PLS-DA method because it is a strictly multi-way classification method and can make the utmost use of the EEM fluorescence information of the vinegar samples. The N-PLS-DA model constructed from the EEM spectra enabled the recognition of adulterated samples with accurate rate of 100% both in training and prediction sets and obtained maximum sensitivity and specificity. This study showed that EEM fluorescence spectroscopy combined with multi-way classification algorithm N-PLS-DA could be used as a good method for the identification of the adulteration of Shanxi aged vinegar samples.

Identification, Characterization, and Prediction of Traffic Flow Patterns in Multi-Airport Systems

Efficient planning of airport capacity is key for the successful accomplishment of traffic flow management. Yet, the dynamic and uncertain behavior of capacity-determining factors makes it difficult to estimate flow rates precisely, especially for strategic planning horizons. Metroplex systems impose additional challenges in this decision-making process because of relevant operational interdependencies between the closely located airports. This paper presents a data-driven framework to identify, characterize, and predict traffic flow patterns in the terminal area of multi-airport systems toward improved capacity planning decision support in complex airspace. Through the identification and characterization of patterns in the terminal area traffic flows, we learn recurrent utilization patterns of runways and airspace as well as relevant decision factors, and use that knowledge to develop descriptive models for metroplex configuration prediction and capacity estimation. The framework is based on the application of machine learning methods on historical flight tracks, weather forecasts, and airport operational data. A multi-layer clustering analysis is first performed to mine spatial and temporal trends in flight trajectory data for identification of traffic flow patterns. Based on this knowledge, a multi-way classification model is developed to generate probabilistic forecasts of the metroplex traffic flow structure for look-ahead times of up to eight hours. Finally, an empirical approach for arrival capacity estimation is proposed based on historical flow pattern behavior. The observed variability in throughput and terminal area delay performance emphasizes the importance of metroplex configuration predictability toward improved flow rate planning and ultimately better traffic regulation.

Front-Face Fluorescence Combined with Second-Order Multiway Classification, Based on Polyphenol and Chlorophyll Compounds, for Virgin Olive Oil Monitoring Under Different Photo- and Thermal-Oxidation Procedures

Room temperature excitation-emission matrices (EEMs), recorded using front-face fluorescence spectroscopy, and combined with second-order multiway classification methods, were explored as potential tools for controlling the storage conditions of extra virgin olive oil samples (EVOO). Factors such as UV-irradiation, sunlight exposition and temperature up to 80 °C were studied. For each sample, EEMs were obtained in two spectral regions, corresponding to polyphenols, and chlorophyll and derivatives, respectively. The full fluorescence information of excitation-emission matrices was processed with parallel factor analysis (PARAFAC) and PARAFAC supervised by linear discriminant analysis (LDA). The models allowed the discrimination between non-irradiated and irradiated EVOO samples, in both spectral regions. With a temperature of 80 °C, and a heating time of 30 min, the formation of secondary oxidation products was appreciable. In these conditions, the first component of PARAFAC showed a remarkable modification in its profile and LDA-PARAFAC allowed the discrimination between non-heated and heated EVOO samples.

EmoSense: Automatically Sensing Emotions From Speech By Multi-way Classification.

Reliably detecting emotions is a topic of current research in understanding mental health. Among the many modes of detecting emotion, audio has a prominent place. In this paper, we propose a two-level, multi-way classifier applied to classification of seven emotions from the standard Emo-DB database. The multi-way classifier is an automated methodology of analyzing a confusion matrix of a first-level classifier to build more classifiers at the next level. A random forest classifier is used on state-of-the-art features for analyzing affective speech. The confusion matrix from this classification level is analyzed to decide, for each class, which other classes are most confused by using a threshold on the misclassification rate. For the chosen pairs, second level classifiers are built and trained on the same data. Its performance on the training-set (73.3{\%) as well as a non-intersecting training set (72.9{\%) are both better than state-of-the-art performance. We initiate a possible explanation of the performance improvement by considering the confusion among emotions placed on Russel's circumplex model.

Sparse-TDA: Sparse Realization of Topological Data Analysis for Multi-Way Classification

Topological data analysis (TDA) has emerged as one of the most promising techniques to reconstruct the unknown shapes of high-dimensional spaces from observed data samples. TDA, thus, yields key shape descriptors in the form of persistent topological features that can be used for any supervised or unsupervised learning task, including multi-way classification. Sparse sampling, on the other hand, provides a highly efficient technique to reconstruct signals in the spatial-temporal domain from just a few carefully-chosen samples. Here, we present a new method, referred to as the Sparse-TDA algorithm, that combines favorable aspects of the two techniques. This combination is realized by selecting an optimal set of sparse pixel samples from the persistent features generated by a vector-based TDA algorithm. These sparse samples are selected from a low-rank matrix representation of persistent features using QR pivoting. We show that the Sparse-TDA method demonstrates promising performance on three benchmark problems related to human posture recognition and image texture classification.

SigMat: a classification scheme for gene signature matching

MotivationSeveral large-scale efforts have been made to collect gene expression signatures from a variety of biological conditions, such as response of cell lines to treatment with drugs, or tumor samples with different characteristics. These gene signature collections are utilized through bioinformatics tools for ‘signature matching’, whereby a researcher studying an expression profile can identify previously cataloged biological conditions most related to their profile. Signature matching tools typically retrieve from the collection the signature that has highest similarity to the user-provided profile. Alternatively, classification models may be applied where each biological condition in the signature collection is a class label; however, such models are trained on the collection of available signatures and may not generalize to the novel cellular context or cell line of the researcher’s expression profile.ResultsWe present an advanced multi-way classification algorithm for signature matching, called SigMat, that is trained on a large signature collection from a well-studied cellular context, but can also classify signatures from other cell types by relying on an additional, small collection of signatures representing the target cell type. It uses these ‘tuning data’ to learn two additional parameters that help adapt its predictions for other cellular contexts. SigMat outperforms other similarity scores and classification methods in identifying the correct label of a query expression profile from as many as 244 or 500 candidate classes (drug treatments) cataloged by the LINCS L1000 project. SigMat retains its high accuracy in cross-cell line applications even when the amount of tuning data is severely limited.Availability and implementationSigMat is available on GitHub at https://github.com/JinfengXiao/SigMat.Supplementary information Supplementary data are available at Bioinformatics online.

An ensemble learning system for a 4-way classification of Alzheimer’s disease and mild cognitive impairment

Discriminating Alzheimer’s disease (AD) from its prodromal form, mild cognitive impairment (MCI), is a significant clinical problem that may facilitate early diagnosis and intervention, in which a more challenging issue is to classify MCI subtypes, i.e., those who eventually convert to AD (cMCI) versus those who do not (MCI). To solve this difficult 4-way classification problem (AD, MCI, cMCI and healthy controls), a competition was hosted by Kaggle to invite the scientific community to apply their machine learning approaches on pre-processed sets of T1-weighted magnetic resonance images (MRI) data and the demographic information from the international Alzheimer’s disease neuroimaging initiative (ADNI) database. This paper summarizes our competition results. We first proposed a hierarchical process by turning the 4-way classification into five binary classification problems. A new feature selection technology based on relative importance was also proposed, aiming to identify a more informative and concise subset from 426 sMRI morphometric and 3 demographic features, to ensure each binary classifier to achieve its highest accuracy. As a result, about 2% of the original features were selected to build a new feature space, which can achieve the final four-way classification with a 54.38% accuracy on testing data through hierarchical grouping, higher than several alternative methods in comparison. More importantly, the selected discriminative features such as hippocampal volume, parahippocampal surface area, and medial orbitofrontal thickness, etc. as well as the MMSE score, are reasonable and consistent with those reported in AD/MCI deficits. In summary, the proposed method provides a new framework for multi-way classification using hierarchical grouping and precise feature selection.

Diversity of Maize Kernels from a Breeding Program for Protein Quality III: Ionome Profiling

Densities of single and multiple macro- and micronutrients were estimated in the mature kernels of 1348 accessions in 13 maize genotypes. The germplasm belonged to stiff stalk (SS) and non-stiff stalk (NS) heterotic groups (HGs) with one (S1) to four (S4) years of inbreeding (IB), or open pollination (OP), and with opaque or translucent endosperm (OE and TE, respectively). Indices were calculated for macronutrients (M-Index), micronutrients (m-Index) and an index based on Fe and Zn densities (FeZn-Index). The objectives were to (1) build predictive models and quantify multivariate relationships between single and multiple nutrients with physical and biochemical constituents of the maize kernel; (2) quantify the effects of IB stages and endosperm textures, in relation to carbon and nitrogen allocation, on nutrients and their indices; and (3) develop and test the utility of hierarchical multi-way classification of nutrients with kernel color space coordinates. Differences among genotypes and among IB stages accounted for the largest amount of variation in most nutrients and in all indices, while genotypic response to IB within HGs explained 52.4, 55.9, and 76.0% of variation in the M-Index, m-Index, and FeZn-Index, respectively. Differences in C and N allocation among HGs explained more variation in all indices than respective differences in allocation among endosperm (E) textures, while variation decreased with sequential inbreeding compared to OP germplasm. Specific color space coordinates indicated either large macronutrient densities and M-Index, or large micronutrient densities, m-Index, and FeZn-Index. These results demonstrated the importance of genotypes and the C:N ratio in nutrient allocation, as well as bivariate and multiple interrelationships.

Detection and quantification of extra virgin olive oil adulteration by means of autofluorescence excitation-emission profiles combined with multi-way classification

Within olive oils, extra virgin olive oil is the highest quality and, in consequence, the most expensive one. Because of that, it is common that some merchants attempt to take economic advantage by mixing it up with other less expensive oils, like olive oil or olive pomace oil. In consequence, the characterization and authentication of extra virgin olive oils is a subject of great interest, both for industry and consumers. This paper reports the potential of front-face total fluorescence spectroscopy combined with second-order chemometric methods for the detection of extra virgin olive oils adulteration with other olive oils. Excitation-emission matrices (EEMs) of extra virgin olive oils and extra virgin olive oils adulterated with olive oils or with olive pomace oils were recorded using front-face fluorescence spectroscopy. The full information content in these fluorescence images was analyzed with the aid of unsupervised parallel factor analysis (PARAFAC), PARAFAC supervised by linear discriminant analysis (LDA-PARAFAC), and discriminant unfolded partial least-squares (DA-UPLS). The discriminant ability of LDA-PARAFAC was studied through the tridimensional plots of the canonical vectors, defining a surface separating the established categories. For DA-UPLS, the discriminant ability was established through the bidimensional plots of predicted values of calibration and validation samples, in order to assign each sample to a given class. The models demonstrated the possibility of detecting adulterations of extra virgin olive oils with percentages of around 15% and 3% of olive and olive pomace oils, respectively. Also, UPLS regression was used to quantify the adulteration level of extra virgin olive oils with olive oils or with olive pomace oils.

Tensor-Factorized Neural Networks.

The growing interests in multiway data analysis and deep learning have drawn tensor factorization (TF) and neural network (NN) as the crucial topics. Conventionally, the NN model is estimated from a set of one-way observations. Such a vectorized NN is not generalized for learning the representation from multiway observations. The classification performance using vectorized NN is constrained, because the temporal or spatial information in neighboring ways is disregarded. More parameters are required to learn the complicated data structure. This paper presents a new tensor-factorized NN (TFNN), which tightly integrates TF and NN for multiway feature extraction and classification under a unified discriminative objective. This TFNN is seen as a generalized NN, where the affine transformation in an NN is replaced by the multilinear and multiway factorization for tensor-based NN. The multiway information is preserved through layerwise factorization. Tucker decomposition and nonlinear activation are performed in each hidden layer. The tensor-factorized error backpropagation is developed to train TFNN with the limited parameter size and computation time. This TFNN can be further extended to realize the convolutional TFNN (CTFNN) by looking at small subtensors through the factorized convolution. Experiments on real-world classification tasks demonstrate that TFNN and CTFNN attain substantial improvement when compared with an NN and a convolutional NN, respectively.

On measure of second-order marginal symmetry for multi-way classifications

On measure of second-order marginal symmetry for multi-way classifications