Maximum Discrimination Research Articles

Preserving empirical data utility in [formula omitted]-anonymous microaggregation via linear discriminant analysis

Today’s countless benefits of exploiting data come with a hefty price in terms of privacy. k-Anonymous microaggregation is a powerful technique devoted to revealing useful demographic information of microgroups of people, whilst protecting the privacy of individuals therein. Evidently, the inherent distortion of data results in the degradation of its utility. This work proposes and analyzes an anonymization method that draws upon the technique of linear discriminant analysis (LDA), with the aim of preserving the empirical utility of data. Further, this utility is measured as the accuracy of a machine learning model trained on the microaggregated data. By transforming the original data records to a different data space, LDA enables k-anonymous microaggregation to build microcells more tailored to an intrinsic classification threshold. To do this, first, data is rotated (projected) towards the direction of maximum discrimination and, second, scaled in this direction by a factor α that penalizes distortion across the classification threshold. The upshot is that thinner cells are built along the threshold, which ends up preserving data utility in terms of the accuracy of machine learned models for a number of standardized data sets.

Evaluation of signs of inflammation that can kill score in predicting mortality in an urban tertiary care centre for children in South India

Background: Mortality in critically ill children is maximum in the first 24 hours. Need of the hour is a rapid clinical scoring system that predicts mortality on admission. This study aims at using the Signs of Inflammation that Can Kill (SICK) score in a tertiary care pediatric hospital in India, to evaluate its usefulness in predicting mortality on admission.Methods: The study design is a hospital based prospective study for evaluation of diagnostic test for a duration of one year. The assessment using SICK score was done on arrival prior to initiation of treatment for children admitted through the emergency room and paediatric ward. The variables of SICK score - temperature, heart rate, respiratory rate, systolic BP, oxygen saturation, CRT, sensorium by AVPU scale were measured using standard guidelines.Results: The study includes a population of 369 children. Children between 1 month to 12 years were included in the study. Out of 369 children studied 24 died. The mortality in the study is 6.5%. The area under the ROC curve is 0.94, which indicates the scores based on regression could predict mortality in 94% subjects correctly. Further a score of 2.5 showed maximum discrimination with a sensitivity of 87.5% and specificity of 87.2%.Conclusions: The assessment of SICK score in the population will provide objective measure of severity of illness on admission, prediction of mortality, early triage of patients, effective allocation of resources and personnel, enables early intervention, which helps in reducing mortality.

Effects of the Conductive Component of Hearing Loss on Speech Discrimination Ability

Effects of decreasing auditory activity on speech discrimination ability are not fully understood. To investigate influence of decrease in auditory activity due to conductive and sensorineural components of hearing loss (HL) on speech discrimination ability. We retrospectively reviewed medical records of patients with suspected HL at Kitasato University Hospital in 2017 and 2018. Patients were divided according to pure-tone audiometry findings: no HL (N-HL), conductive HL (C-HL), sensorineural HL (S-HL), and mixed HL (M-HL) groups. In total, 149 patients (224 ears) were eligible. The maximum speech discrimination score (SDSmax) for all ears significantly negatively correlated with age (r = -0.29, p<0.0001) and bone conduction (BC) threshold (r = -0.55, p<0.0001). For patients aged <50 years in N-HL and C-HL groups, SDSmax was nearly 100%, with no significant difference. SDSmax was significantly lower for older patients (≥50 years) in the M-HL group than in the S-HL group, even though there were no significant differences in age and BC thresholds between groups. Decrease of auditory activity due to the conductive component of M-HL may worsen speech discrimination ability. Early treatment of M-HL would be desirable for the preservation of auditory function.

Cochlear implantation in patient with Charcot-Marie-Tooth disease

Two patients with auditory neuropathy spectrum disorder (ANSD) considered to be associated with Charcot-Marie-Tooth (CMT) are reported. In case 1, a 23-year-old man presented with progressive bilateral sensorineural hearing loss (SNHL) from 10 years of age and was diagnosed with ANSD. He was later diagnosed with CMT by neurological testing. In case 2, a 16-year-old girl, the younger sister of the patient in case 1, presented with progressive SNHL with similar auditory findings since 6 years of age as those of her brother. Both cases underwent bilateral cochlear implantation. In case 1, the maximum discrimination score improved to 45% 24 months after the first side of cochlear implantation from 5% before the surgery. In case 2, the score was 5% 10 months after cochlear implant (CI) surgery from 0% before the surgery. CI treatment for CMT patients has been considered difficult because of both failure in synchronization of nerve conduction due to demyelination and axonal failure of the auditory nerve. Though slower progress compared to the average subset of patients receiving CI was seen, significant improvement was gradually observed in the present patients after bilateral cochlear implantations. CI is thus a viable option for rehabilitation of SNHL in CMT patients.

Maximum discrimination index: a tool for land cover identification

This work presents an adaptable index that is applied to a pair of covers to be discriminated. Its adaptability relies on the procedure to determine the numerical value of the wavelengths or bands involved: the maximization of an operator based on the geometric mean of squared differences. This index is applied to the particular case of discrimination of wheat from ryegrass in different phenological stages. The maximum discrimination index outperforms other indices such as the normalized difference vegetation index, advanced normalized vegetation index and normalized difference greenness index. Its efficacy of discrimination is characterized and compared with the normalized difference greenness index (the second with better performance). It is observed that the proposed index has a more predictable behavior and reaches a discrimination accuracy as high as 95.5%. The maximum discrimination index could be adjusted to different covers and employed as a tool for discrimination. Spectral signatures coming from any platform: field, aerial or satellite, can be handled.

Adaptive appearance modeling via hierarchical entropy analysis over multi-type features

The descriptiveness of visual models is crucial for many image processing applications, however, it is still challenging to adaptively formulate such models. This paper systematically advocates a generic and adaptive appearance modeling method. For object-specific instances in images, it can adaptively generate a descriptive codebook by exploring the maximum discriminability of multi-type features. The key idea is to define feature-independent information entropy as a unified criterion to measure different features in a common entropy space. Towards this goal, a hierarchical maximum entropy (HME) model is proposed to conduct multi-feature selection based on the random forest. Specifically, the improved random forest replaces space-specific expression “distance similarity” with the statistical concept “entropy”. Thus, the random forest could integrate the subspace clustering results from different feature spaces. Such integration can not only afford adaptive feature selection and cross-feature error control but also be robust to possible weak/inconsistent feature expressions. To effectively construct a class-specific appearance model, a sparse codebook model, consisting of a series of weak learners, is proposed to further explore the maximum discriminative subspaces of each object class. Finally, a maximum entropy model is proposed to formulate appearance model by optimizing the probabilistic distributions of all the codebook words’ response parameters. To verify the efficacy and effectiveness of the proposed model, it is applied to multi-class image classification. We conduct extensive experiments and make comprehensive evaluations w.r.t several state-of-the-art methods over PASCAL VOC 2007, VOC 2012, Caltech 101 and Caltech 256 datasets. All the results demonstrate the advantages of the our method in terms of precision, robustness, flexibility, and versatility.

Increasing the discrimination power of rapid evaporative ionisation mass spectrometry (REIMS) in analytical control tissue quality screening and cell line sample identification.

Rapid Evaporative Ionisation Mass Spectrometry (REIMS) has been evaluated as a tool to improve analytical efficiency and add capability in areas within Pharmaceutical Research and Development (Pharma R&D). This article reports the comparison of single MS, and tandem MS/MS REIMS (REIMS and REIMS/MS) methodologies to investigate which mode produces maximum discrimination power for screening applications. Control tissue samples and cell line suspension samples were analysed using optimised REIMS and REIMS/MS to evaluate which technique produced optimal discrimination power for control tissue and cell line identification. The iKnife sampling tool and a prototype 'cell sampler' were utilised for tissue and cell analysis, respectively. The REIMS source was coupled to a hybrid Quadrupole-Time Of Flight (QTOF) mass spectrometer. Multivariate Analysis (MVA) was utilised to evaluate the resulting Mass Spectrometry (MS) data and discriminate between sample types. Proof of concept investigations demonstrating that REIMS/MS offered increased MVA discrimination for sample identification, compared with REIMS, is presented for the first time. Control tissue data showed discrimination by timepoint classification over 0-144 h storage after removal from the host. Timepoint discrimination was optimised using REIMS/MS with a collision energy that effectively maximised ion fragmentation. Similar optimisation was observed when REIMS/MS was applied to the identification of cell lines. The proof of concept results demonstrate that REIMS/MS can offer advantages over REIMS for control tissue quality screening, and cell line identification applications in Pharma R&D. Further work following this proof of concept investigation is being undertaken to implement the technology for these applications, utilising the optimised REIMS/MS methodology. REIMS/MS will also be used as an optimised tool for other applications.

A Novel Credit Evaluation Model Based on the Maximum Discrimination of Evaluation Results

ABSTRACT This paper proposes a novel model for establishing a credit evaluation system, including a system of indicators, indicator weights, and credit scores. A credit evaluation system whose evaluation results have significant discrimination is good. Based on this standard, we construct an objective programming model with the maximum discrimination of credit scores as the objective function. The main constraint condition is that the indicator weights sum to 1, and weight is a decision variable. After we delete indicators whose weight is 0, we design a system of indicators, and then obtain credit scores with the maximum discriminatory power. Our empirical study of China’s 3,045 small businesses confirms that this model is both easy to use and reasonable. The empirical results show that, compared to logistic regression and CHAID decision trees, our model has greater accuracy based on F, AUC, and KS tests.

The Role of Phonotactic Information in the Processing and Production of Inflectional Morphology

In two separate studies, it was examined to what extent in languages with rich inflectional morphology, such as Serbian, in the tasks of automatic perception and word production we can rely on phonotactic information, i.e. permitted combinations of phonemes/graphemes. In the first study, with the help of support vector machines, word classes discrimination was carried out based on bigrams and trigrams generated at the morphosyntactic category level. In the second study, the production of inflected forms was explored, with the help of memory-based learning, relying on phonotactic information from the last four syllables of lemmas. Maximum discrimination accuracy of inflected word classes was obtained on the basis of bigrams reaching about 93% of accuracy. Similarly, in the task of inflectional production on all inflected word classes taken together, around 92% of the inflected forms were generated correctly. This confirms that phonotactic information in the perception and production of morphologically complex words plays an important role. Therefore, this information should be taken into account when considering the emergence of larger language units and patterns. The results show that functional connections between orthography/phonology and semantics lead to successful lexical learning, but at the same time they call into question the need for positing the existence of mental lexicon in which mental representations of various language characteristics are stored.

A supplier performance evaluation framework using single and bi-objective DEA efficiency modelling approach: individual and cross-efficiency perspective

In view of complexities associated with supplier performance evaluation based on traditional business criterions (such as costs, quality levels, and delivery timelines) and emerging criterions (such as those related to environmental sustainability), we in this research evolve two different supplier efficiency measurement models that unify such criterions possessing characteristics of both desirable and undesirable outputs. The first model is a single-objective DEA efficiency assessment model wherein both types of outputs are integrated into a single composite efficiency measure. Using data from suppliers of Hyundai Steel Company, we determine composite efficiencies of each of these suppliers thus ranking them in terms of an overall efficiency score that would be useful as far as the first-cut supplier discrimination is concerned. However, due to the relative inability of evolved single-objective efficiency model to perform trade-offs amongst desirable and undesirable outputs and, owing to unidimensionality aspects, we evolve a goal programming based bi-objective efficiency model wherein trade-offs can be performed between both conventional and emerging dimensions criterions leading to different supplier evaluations for varied scenarios. We also integrate our evolved models with the cross-efficiency view of efficiency determination in order to enable the decision-makers to achieve peer-to-peer evaluation and maximum discrimination amongst suppliers.

The decision tool unipolar depression (DTUD): a new measure to facilitate the early identification of patients with major depressive disorder in need of highly specialized care

BackgroundSelection of the optimal initial treatment in patients with major depressive disorder (MDD) in need of highly specialized care has the potential to benefit treatment outcomes and cost-effectiveness of treatment strategies. However, to date, there is a paucity of measures that could guide the selection of the initial treatment, in particular to indicate which patients with MDD are in need of highly specialized care. Recognizing this gap, this paper reports on the development and psychometric evaluation of the Decision Tool Unipolar Depression (DTUD), aimed to facilitate the early identification of patients with MDD in need of highly specialized care.MethodsThe DTUD was developed using a mixed-methods approach, consisting of a systematic review and a concept mapping study. To evaluate the psychometric features of the DTUD, a cross-sectional multicenter study was conducted. A total of 243 patients with MDD were evaluated with the DTUD. Feasibility was operationalized as the time required to complete the DTUD and the content clarity of the DTUD. Inter-rater reliability was evaluated using Krippendorf’s alpha. The Maudsley Staging Method (MSM) and the Dutch Measure for quantification of Treatment Resistance in Depression (DM-TRD) were administered to assess the convergent validity. A receiver operator characteristic curve was generated to evaluate the criterion validity and establish the optimal cut-off value.ResultsThe mean administration time was 4.49 min (SD = 2.71), and the content of the total DTUD was judged as clear in 94.7% of the evaluations. Inter-rater reliability values ranged from 0.69 to 0.91. Higher scores on the DTUD were associated with higher scores on the MSM (rs = 0.47) and DM-TRD (rs = 0.53). Based on the maximum Youden index (0.494), maximum discrimination was reached at a cut-off score of ≥5 (sensitivity 67%, specificity 83%).ConclusionThe DTUD demonstrated to be a tool with solid psychometric properties and, therefore, is a promising measure for the early identification of patients with MDD in need of highly specialized care. Use of the DTUD has the potential to facilitate the selection and initiation of the optimal initial treatment in patients with MDD, which in turn may improve the clinical effectiveness and cost-effectiveness of treatment strategies.

Diagnostic Accuracy of Monocyte Chemotactic Protein (MCP)-2 as Biomarker in Response to PE35/PPE68 Proteins: A Promising Diagnostic Method for the Discrimination of Active and Latent Tuberculosis.

Several studies have been conducted to find new biomarkers for the discrimination of Latent Tuberculosis Infection (LTBI) from active TB (ATB); however, their findings are inconsistent. The aim of the current study was to evaluate the potential of in vitro antigenspecific expression of Monocyte Chemotactic Protein (MCP)-2 for discrimination of ATB and LTBI after stimulation of whole blood with PE35 and PPE68 recombinant proteins. The recombinant PE35 and PPE68 proteins were evaluated at a final concentration of 5 µg/ml by a 3-day whole blood assay. Secreted MCP-2 from the culture supernatants were measured by commercially available Human MCP2 ELISA Kit. The diagnostic performance of MCP-2 was ascertained by Receiver Operator Characteristic (ROC) curve and measuring the Area Under the Curve (AUC) and their 95% Confidence Intervals (CI). Cut-offs was estimated at various sensitivities and specificities and at the maximum Youden's index (YI), i.e. sensitivity specificity-1. The median MCP-2 response to both PE35 and PPE68 in those with LTBI was significantly higher than patients with ATB. The discrimination performance of MCP-2 response following stimulation of PE35 (assessed by AUC) between LTBI and patients with ATB was 0.98 (95%CI: 0.94-1.00). Maximum discrimination was reached at a cut-off of 86pg/mL with 100% sensitivity and 97% specificity. The highest sensitivity and specificity was obtained using cut off 58 pg/mL following stimulation with PPE68 (100% and 90%, respectively; AUC: 0.94, 95%CI: 0.85- 1.00). MCP-2 induced by PE35 and PPE68 shows good discriminatory power for discrimination of ATB and LTBI. Additional studies with a larger sample size are needed to confirm the advantage of this marker, alone or combined with other markers; however, these findings present a promising method, which can discriminate between ATB and LTBI.

Four-Sequence Maximum Entropy Discrimination Algorithm for Glioma Grading

Grading of glioma is crucial for treatment decision making as well as prognostic assessments. In clinical routines, radiologists grade gliomas with multiple complementary magnetic resonance imaging (MRI) sequences, which is yet challenging for glioma prediction models. In this paper, we take full advantages of four commonly used MRI sequences to propose non-invasive grading of glioma based on a variant of maximum entropy discrimination (MED) and decision tree. First, radiomics features calculation is, respectively, performed on T1-weighted imaging, T2-weighted imaging, fluid attenuation inversion recovery imaging, and contrast-enhanced T1-weighted imaging. Then, radiomics features are integrated to build a glioma prediction model named four-sequence MED (FSMED) according to the assumption that the classification margin of different sequences is consistent. Finally, we propose a multi-MED decision tree (MMEDT) model to obtain the grading of gliomas based on the output of FSMED and the results of MED on each sequence. Validation experiments are conducted on a data set collected from Henan Provincial People's Hospital (GliomaHPPH2018) and Multimodal Brain Tumor Image Segmentation Benchmark 2017 (BraTS2017). The results of these two data sets demonstrate the high-prediction performance of our method. The average areas under the curve (AUC) of MMEDT are 0.9119, 0.8184, and 0.9084 for GliomaHPPH2018, BraTS2017, and their merged set, respectively, with the corresponding average sensitivities of 92.55%, 87.85%, and 87.91%, and average specificities of 92.57%, 81.36%, and 87.39%.

Категоризация этнических лиц русскими и тувинцами в дискриминационных задачах

The opposite tendencies towards blurring the boundaries of compact living of ethnic groups and of relocation of large groups of strangers into the territories of compact residence of local groups generate a number of specific phenomena connected with the inter-group perception. We conducted a comparative study of perception and communication of transition series of “ethnic faces” among two small-scale isolated populations and representatives of the corresponding ethnic groups in multi-ethnic cities. The participants were to solve two discrimination task, one individually, the AB=X task, the other one, the dual “same-different” task, in dyads. For all samples of subjects, of In the AB=X task, the maximum discrimination efficiency was recorded for the same no. 2—3 pair of transition images series This may indicate that all subjects relied on the morphological features stimulus faces than on anything else. Also a tendency is observed toward similarity in the dynamics of efficiency of task performance in different groups of subjects, with the exception of the sample of Russians from the metropolis. At the same time, in the dyadic experiment, subjects from megacities demonstrate different trends in terms of the maximum efficiency of task performance. The sample of Russian subjects from Moscow demonstrates a significantly smaller decrease in efficiency when approaching the edges of the transitional series of stimulus images in the AB=X task, but not in the dual task. This is probably due to different nature of the tasks, i.e. one may easily suppose a more developed ability to perceptually discriminate between ethnic faces resulting from their greater variety in the environment, as in Moscow megacity, whereas it does not seem reasonable to suppose difference in communication skills required to perform the dual task across the samples.

General multi-view learning with maximum entropy discrimination

Maximum entropy discrimination (MED) is an excellent discriminative classification algorithm in view of maximum entropy and maximum margin principles, and can generate hard-margin support vector machines (SVMs) under certain condition. Recently, many versions of multi-view maximum entropy discrimination are proved to be effective in multi-view learning. However, those methods are designed only for the two-view classification problem. For example, alternative multi-view maximum entropy discrimination (AMvMED) and multi-kernel multi-view maximum entropy discrimination (MkMvMED) belong to the two-view classification method. In this paper, we extend AMvMED and MkMvMED to the general multi-view classification problems by jointly learning multiple different views in a non-pairwise way called general alternative multi-view maximum entropy discrimination (GAMvMED) and general multi-kernel multi-view maximum entropy discrimination (GM3ED). In GAMvMED, combination weight for the reconstruction of each view in regularization terms is learned to explore complementarity information among different views. Finally, we provide an efficient iteration algorithm with the classical convex quadratic programming for optimization. Simultaneously, we derive that GAMvMED and GM3ED are equivalent in form under the similar prior assumption. Experimental results confirm the effectiveness of our proposed methods.

Multimodal correlation deep belief networks for multi-view classification

The Restricted Boltzmann machine (RBM) has been proven to be a powerful tool in many specific applications, such as representational learning, document modeling, and many other learning tasks. However, the extensions of the RBM are rarely used in the field of multi-view learning. In this paper, we present a new RBM model based on canonical correlation analysis, named as the correlation RBM, for multi-view learning. The correlation RBM computes multiple representations by regularizing the marginal likelihood function with the consistency among representations from different views. In addition, the multimodal deep model can obtain a unified representation that fuses multiple representations together. Therefore, we stack the correlation RBM to create the correlation deep belief network (DBN), and then propose the multimodal correlation DBN for learning multi-view data representations. Contrasting with existing multi-view classification methods, such as multi-view Gaussian process with posterior consistency (MvGP) and consensus and complementarity based maximum entropy discrimination (MED-2C), the correlation RBM and the multimodal correlation DBN have achieved satisfactory results on two-class and multi-class classification datasets. Experimental results show that correlation RBM and the multimodal correlation DBN are effective learning algorithms.

Quantification of Degree of Liver Fibrosis Using Fibrosis Area Fraction Based on Statistical Chi-Square Analysis of Heterogeneity of Liver Tissue Texture on Routine Ultrasound Images

We present a novel method to quantify the degree of liver fibrosis using fibrosis area fraction based on statistical chi-square analysis of heterogeneity of echo texture within liver on routine ultrasound images. We demonstrate, in a clinical study, that fibrosis area fraction derived this way has the potential to become a noninvasive, quantitative radiometric discriminator of normal or low-grade liver fibrosis (Ishak fibrosis score range = F0-3) and advanced liver fibrosis or cirrhosis (Ishak fibrosis score range = F4-6) on routine ultrasound images. This retrospective patient study was institutional review board approved. Ultrasound images of 100 patients (61males, 39 females; 18-81 years) who underwent nontargeted ultrasound-guided biopsy were randomly divided into two groups: a training group consisted of 31 cases, and a validation group that contained the rest cases. An investigator manually selected a primary region of interest (ROI; approximately 4-6 cm2) in the liver tissue while avoiding nonhepatic parenchyma. The primary ROI contained a large number of secondary ROIs (25 × 25 pixels) to maintain the precision of statistical analysis. Sample variance σ2 of image gradient (a texture feature related to the amount of edge structures) was calculated in secondary ROIs in a roster scan fashion. A theoretical derivation was presented to estimate population variance [Formula: see text] of image gradient across the primary ROI from mean gradient µ of secondary ROIs. The χ2 (χ2 = σ2/ [Formula: see text] ) was computed at each secondary ROI, forming a χ2 map of liver tissue heterogeneity. A cut-off value was optimized from datasets in the training group by comparing to the fibrosis grades determined by biopsy. This cut-off value was then applied to the datasets in the validation group to convert the χ2 maps into binary images, from which fibrosis area fractions (fraction of fibrosis area to the total area of the primary ROI) were calculated and entered in a statistical analysis. In the training group, the optimal setting was found to be [Formula: see text] = 6.0, which resulted a maximum discrimination of F0-3 vs F4-6: p < 0.0001, area under curve = 0.985, sensitivity = 93.7%, specificity = 93.3%. When this setting was applied to the datasets in the validation group, a distinct separation was seen between the two classes (p < 0.0001). F0-3 class had an average fibrosis area fraction of 4.7% (1.7%-11.4%), whereas the F4-6 class had an average fibrosis area fraction of 17.3% (9.8%-29.6%). A strong correlation was demonstrated between the fibrosis area fraction and histological fibrosis grade determined by biopsy (area under curve = 0.89, sensitivity = 87.9%, specificity = 90.3%). The presented method is a promising noninvasive tool for distinguishing normal or low-grade liver fibrosis (F0-3) and advanced liver fibrosis or cirrhosis (F4-6) from routine ultrasound images. These findings support the further development of texture heterogeneity analysis, particularly fibrosis area fraction, as a quantitative biomarker for distinguishing various liver fibrosis grades.

Effect of Binaural Hearing Aids on the Improvement of the Maximum Discrimination Score

両耳聴の意義に対する認識が高まり, 補聴器も人工内耳も可能であれば両耳聴を目指す方向に向かっている. 今回, 装用前の語音弁別能は不良であったが, 両耳装用後に実用レベルに達した症例について検討した. 耳鳴を主訴とした両側水平性感音難聴の2症例で, 語音弁別能は初診時左右それぞれ30～55%であったが, 1年後には両耳装用下の音場検査にて各々80,95%という結果が得られた. これは周波数ごとに圧縮を変更しながら十分なファンクショナルゲインを確保したことと, 優れた両耳聴効果に起因するところが大きいと思われる. そして, 補聴器の調整を適正に行い十分な聴能訓練を行うことで, 補聴器の効果をより発揮できると考える.

Multi-view learning for visual violence recognition with maximum entropy discrimination and deep features

Harmful information full of violence, whose transmission is difficult to control, is one of the negative effects that the Internet brings to us today. Such corrosive information in the network environment infiltrates the mental health of adolescents. Many young people who lack judgment tend to imitate violent behaviors that are propagated on the network. This study applies multi-view learning to violent behavior recognition of still images, and can be useful for the research of network image or video information monitoring and filtering. In this paper, to start with, we construct a violence image recognition database, which contains 5974 violence images and 11,516 non-violence images. From this database, different datasets can be constructed for model and algorithm evaluations. Then we propose a multi-view maximum entropy discriminant (MVMED+) model for learning with different numbers of views. It can combine various features of the images and thus classify the images with the complementary information between the views. For efficient optimization, we further derive a sequential minimal optimization algorithm to train the model. In the experiments, we use handcrafted features and deep learning features to validate the effectiveness of the proposed model, respectively. As a useful byproduct, large performance improvements on visual violence recognition are further observed with deep learning features.

Potential role of nuclear magnetic resonance spectroscopy to identify salivary metabolite alterations in patients with head and neck cancer.

The analysis of the salivary metabolomic profile may offer an early phase approach to assess the changes associated with a wide range of diseases including head and neck cancer. The aim of the present study was to investigate the potential of nuclear magnetic resonance (NMR) spectroscopy for detecting the salivary metabolic changes associated with head and neck squamous cell carcinoma (HNSCC). Unstimulated whole-mouth saliva samples collected from HNSCC patients (primary tumour was located either in the larynx or in the oral cavity) and healthy controls were analysed by 1H-NMR spectroscopy. Reliably identified salivary metabolites were quantified and the determined concentration values were compared group-wise using a Mann-Whitney U-test. Multivariate discrimination function analysis (DFA) was conducted to identify such a combination of metabolites, when considered together, that gives maximum discrimination between the groups. HNSCC patients exhibited significantly increased concentrations of 1,2-propanediol (P=0.032) and fucose (P=0.003), while proline levels were significantly decreased (P=0.043). In the DFA model, the most powerful discrimination was achieved when fucose, glycine, methanol and proline were considered as combined biomarkers, resulting in a correct classification rate of 92.1%, sensitivity of 87.5% and specificity of 93.3%. To conclude, NMR spectrometric analysis was revealed to be a feasible approach to study the metabolome of saliva that is sensitive to metabolic changes in HNSCC and straightforward to collect in a non-invasive manner. Salivary fucose was of particular interest and therefore, controlled longitudinal studies are required to assess its clinical relevance as a diagnostic biomarker in HNSCC.