When the cut-off value for a continuous diagnostic variable is increased (assuming that larger values indicate an increased chance of a positive outcome), the proportions of both true and false positives decreases. These proportions are the sensitivity and 1 – specificity, respectively. A graph of sensitivity against 1 – specificity is called a receiveroperatingcharacteristic (ROC) curve. Figure 1 shows the ROC curve for lactate using the cut-off values given in Table 4. The preferred method is to join the points by straight lines but it is possible to fit a smooth curve from a parametric model. A perfect test would have sensitivity and specificity both equal to 1. If a cut-off value existed to produce such a test, then the sensitivity would be 1 for any non-zero values of 1 – specificity. The ROC curve would start at the origin (0,0), go vertically up the y-axis to (0,1) and then horizontally across to (1,1). A good test would be somewhere close to this ideal. If a variable has no diagnostic capability, then a test based on that variable would be equally likely to produce a false positive or a true positive:
In multivariate classification, attention is required for those reference values of markers which provide at least a moderate amount of classification. In usual context of assessing the performance of a test, scores which are nearer to reference value are given the same amount of weightage as that of the scores farther from reference value. The area under the curve (AUC) so computed will be contaminated, and the true accuracy or the actual performance will be masked. This misleads the interpretation of the measures of ROC as well as the optimal threshold. In general, let us consider two tests A and B for better identification of a particular abnormality in individuals. Suppose that the curves of tests A and B cross each other and have at most similar accuracies. Under these circumstances, it is very difficult to notify a better test which has more ability to distinguish status of individuals. To resolve this issue, a new testing procedure is given in a parametric sense which makes use of the modified version of AUC rather than the conventional AUC of MROC curve proposed by Sameera et al. (4). Further, the role of sample size in distinguishing the crossover MROC curves is also considered, and numerical illustrations are accommodated by both real and simulated environments.
RESULTS. For children aged 5 to 18 years, BMI-for-age, triceps skinfold-for-age, and subscapular skinfold-for-age each performed equally well alone in the receiveroperatingcharacteristiccurves in the identification of excess body fat defined by either the 85th or 95th percentile of percentage of body fat by dual-energy radiograph absorptiometry. However, if BMI-for-age was already known and was ⬎ 95th percentile, the additional measurement of skinfolds did not significantly increase the sensitivity or specificity in the identification of excess body fat.
Measuring model performance of rating systems is a major task for banks. The concept of discrimination, i.e. the discriminative power, is used in credit risk modeling to assess the quality of a risk model concerning the separation of extreme events. For PD models CAP (Cumulative Accuracy Profile) or ROC (ReceiverOperatingCharacteristic) curves are used to build a quantity called Accuracy Ratio, which is used to measure the discriminative power. These ideas are well known and broadly used in practice. Although such a measure is also desirable for models of the loss given default (LGD models), it is not documented in the literature. In this note we close this gap. We develop a measure for the discriminative power of LGD models based on Lorenz curves. We study first properties and introduce some alternatives for its cal- culation from a practical point of view.
and known pregnancy outcome between August 2003 and August 2012 were included. Receiveroperatingcharacteristiccurves were used to determine the ideal NT and CRL discordances cut-off points that max- imized the ability to predict adverse outcome, which was defined as any of: death of one or both twins, twin-to-twin transfusion syndrome, or estimated fetal weight or birth weight discordances ≥25%. Results: Of the 89 cases, 20 (22.5%) had at least one adverse outcome. NT discordance was more discriminatory of adverse outcome than was CRL discordance. The optimal values for predicting any adverse outcomes for NT were > 23.7% and for CRL > 3.5%. The positive predictive values for NT (52.4%) and CRL (29.8%) screening were relatively low; however, the lack of either NT or CRL discordances was more reassuring, with negative predictive values of 86.8% and 86.4%, respectively. Conclusions: NT discordance is more predic- tive for adverse fetal outcome in MCDA twins than CRL discordance. Neither NT nor CRL discordance are likely to modify the intensive monitoring required for these very high-risk pregnancies.
Notes: The combined CAGE and GGT scores were calculated by equations derived from multiple logistic regression analyses after adjustment for age and gender. Abbreviations: AUROC, area under the receiveroperatingcharacteristiccurves; NC, normal control; PD, problem drinking; AUD, alcohol use disorder; AD, alcohol dependence; AUDIT, alcohol use disorders identiﬁcation test; CAGE, cutting down, annoyance by criticism, guilty feeling, and eye-openers; GGT, gamma-glutamyl transferase; CAGE+GGT, combination of CAGE and GGT.
obtained via linear regression, controlling for age, sex and height. Validity of SGRQ total score and component categories as predictors of COPD status in the commu- nity was assessed via unadjusted logistic modelling and receiveroperatingcharacteristic (ROC) curves. Based on this outcome, a cut point was chosen and agreement between SGRQ and COPD status using this cut point was established via Fleiss’ kappa test. The association between SGRQ score and COPD severity as defined by per cent predicted FEV 1 was assessed via Pearson correla-
Quantitative imaging has an important role in developing image analysis methods to extract quantitative data more accurately in detection and diagnosis (Giger, 2008). ReceiverOperatingCharacteristic (ROC) analysis is one of the imaging analysis methods to compare the accuracy of two or more imaging modalities. The ROC curve represents sensitivity (True Positive Fraction) and specificity (True Negative Fraction). ROC analysis evaluates the plots by calculating the area under the curves. The mammographic images are scored based on subjective interpretation by observers. Many researchers had proved that quantitative diagnostic assessments from ROC curves could provide valuable feedback to improve diagnostic performance.
FliD and CagA are important virulence factors of H. pylori. We aimed to evaluate the screening values of FliD and CagA for gastric cancer (GC). Serum samples were obtained from 232 cases and 266 controls in a case-control study. Unconditional multivariate logistic regression with odds ratios (ORs) and 95% confidence intervals (CIs) was used to analyze the relationships between FliD, CagA and GC. The sensitivities, specificities and receiveroperatingcharacteristic (ROC) curves were calculated. Finally, the combined screening values of FliD, FlaA, NapA and CagA were assessed based on discriminant analysis. In all subjects, the associations of FliD and CagA with GC were evident with ORs (95% CIs) of 7.6 (4.7-12.3) and 2.5 (1.6-3.8), respectively (*p<0.001). The areas under ROC curves (AUCs) for FliD and CagA were 0.800 and 0.653, respectively. The AUC for the combination of FliD, FlaA and NapA was 0.915, which represented an increase of 0.115 over that of FliD alone (*p<0.001). These findings indicate that the FliD antibody is associated with GC and could exhibit high validity as a biomarker in screening for GC patients. The combination of FliD, FlaA and NapA improved the screening validity.
Continuous data are presented as mean ± standard deviation or median with 25th and 75th percentile, and categorical or ordinal variables are presented as absolute and relative frequencies. On a per patient level, comparisons between groups were performed with an independent student t-test or the Wilcoxon rank-sum test for continuous variables, Fisher’s exact test for categorical variables, and Wilcoxon rank-sum test for ordinal variables. On a segmental level, comparisons between groups were performed using univariable multilevel mixed-effects logistic regression models that account for the group structure (i.e. multiple segments per patient) and thus possible autocorrelation effects in the data. To evaluate interobserver agreement among four readers, intraclass correlation coefficient (ICC) was calculated using two-way random-effects models. To determine co-linearities among all HRP features, we assessed Spearman’s rank correlation coefficients. Uni- and multivariable multilevel mixed-effects logistic regression analysis were performed to identify determinants for culprit lesion within ACS patients and determinants for ACS among all patients. In the multivariable model we included remodeling index, minimal luminal area (mm 2 ), plaque burden (%), TCFA equivalent – Low HU plaque volume <30 HU (mm 3 ) and lesion length. The discriminatory capacity of the dichotomized HRP thresholds for the prediction of culprit lesion and ACS was assessed using the area under curve (AUC) 27 . Multivariable logistic regression models were used to calculate receiveroperatingcharacteristic (ROC) curves. A two-tailed p-value of <0.05 was considered statistically significant. All statistical analyses were performed using Stata 13.1 (Stata- Corp LP, College Station, Texas). The complete source data from the Rule Out Myocardial Infarction/Ischemia Using Computer Assisted Tomography (ROMICAT) II trial are publicly available in accordance with the data sharing policy of the National Institutes of Health and can be used for the purposes of reproducing the results or replicating the procedures.
Along with the developments of deep learning, many recent architectures have been proposed for face recognition and even get close to human performance. However, accurately recognizing an identity from seriously noisy face images still remains a challenge. In this paper, we propose a carefully designed deep neural network coined noise-resistant network (NR-Network) for face recognition under noise. We present a multi-input structure in the final fully connected layer of the proposed NR-Network to extract a multi-scale and more discriminative feature from the input image. Experimental results such as the receiver-operatingcharacteristic (ROC) curves on the AR database injected with different noise types show that the NR-Network is visibly superior to some state-of-the-art feature extraction algorithms and also achieves better performance than two deep benchmark networks for face recognition under noise.
Statistical analysis was performed using analysis of variance to compare mean differences between the groups, as well as receiveroperatingcharacteristic (ROC) curves to determine the specificity and sensitivity of SMR scores to detect PDN, and the coefficient of correlation between the CAN score and SMR score in all the subjects included in the study.