Classification Error Probability Research Articles

On the cost of risk misspecification in insurance pricing

AbstractIn the non-life insurance industry, pricing is often done relative to individual criteria of policyholders. Various classification algorithms are in use to categorize policyholders into risk classes defined by the insurer, but classification errors may result from this process. In the light of recent automatic classification practices, it becomes important to assess the risks caused by such errors. In this paper we examine the impact of risk class misspecifications for a simple situation with two risk types. We provide a mean-variance framework for quantitatively studying the insurer’s optimization problem of specifying premiums and we analyze the tradeoff of costs and benefits when classification error probabilities are known.

Toward a Comprehensive Evaluation of Student Knowledge Assessment for Art Education: A Hybrid Approach by Data Mining and Machine Learning

By analyzing students’ understanding of a certain subject’s knowledge and learning process, and evaluating their learning level, we can formulate students’ learning plans and teachers’ curricula. However, the large amount of data processing consumes a lot of manpower and time resources, which increases the burden on educators. Therefore, this study aims to use a machine learning model to build a model to evaluate students’ learning levels for art education. To improve the prediction accuracy of the model, SVM was adopted as the basic model in this study, and was combined with SSA, ISSA, and KPCA-ISSA algorithms in turn to form a composite model. Through the experimental analysis of prediction accuracy, we found that the prediction accuracy of the KPCA-ISSA-SVMM model reached the highest, at 96.7213%, while that of the SVM model was only 91.8033%. Moreover, by putting the prediction results of the four models into the confusion matrix, it can be found that with an increase in the complexity of the composite model, the probability of classification errors in model prediction gradually decreases. It can be seen from the importance experiment that the students’ achievements in target subjects (PEG) have the greatest influence on the model prediction effect, and the importance score is 9.5958. Therefore, we should pay more attention to this characteristic value when evaluating students’ learning levels.

A class distribution learning method for few-shot remote sensing scene classification

ABSTRACT In recent years, few-shot remote sensing scene classification has become an important foundational research topic, aiming to achieve better classification performance with limited labeled data. Due to the large intra-class variances and inter-class similarity of remote sensing scenes, the scene classification task for remote sensing is much more challenging than general few-shot image classification. In this letter, we propose a class distribution learning method under limited samples by fully using the support samples, which consists of intra-class feature aggregation (IFA) module, intra-class feature homogenization module (IFH) and inter-class cross calibration (ICC) module. IFA and IFH modules are proposed to enable the aggregation and even distribution of intra-class features around the prototypes generated by the model, reducing intra-class distances and mitigating the generation of extreme samples near decision boundaries. ICC module is designed to encourage the features generated by the model for inter-class to be further apart, thereby increasing the inter-class distance and reducing the probability of classification errors. To demonstrate the effectiveness of our proposed method, we conduct comprehensive experiments on three benchmark remote sensing scene datasets: NWPU-RESISC45, WHU-RS19, and UC Merced, achieving extremely advanced performance. Our source codes have been available at https://github.com/mitsuha97/CDLM.

Resolving the non-uniformity in the feature space of age estimation: A deep learning model based on feature clusters of panoramic images

Age estimation is important in forensics, and numerous techniques have been investigated to estimate age based on various parts of the body. Among them, dental tissue is considered reliable for estimating age as it is less influenced by external factors. The advancement in deep learning has led to the development of automatic estimation of age using dental panoramic images. Typically, most of the medical datasets used for model learning are non-uniform in the feature space. This causes the model to be highly influenced by dense feature areas, resulting in adequate estimations; however, relatively poor estimations are observed in other areas. An effective solution to address this issue can be pre-dividing the data by age feature and training each regressor to estimate the age for individual features. In this study, we divide the data based on feature clusters obtained from unsupervised learning. The developed model comprises a classification head and multi-regression head, wherein the former predicts the cluster to which the data belong and the latter estimates the age within the predicted cluster. The visualization results show that the model can focus on a clinically meaningful area in each cluster for estimating age. The proposed model outperforms the models without feature clusters by focusing on the differences within the area. The performance improvement is particularly noticeable in the growth and aging periods. Furthermore, the model can adequately estimate the age even for samples with a high probability of classification error as they are located at the border of two feature clusters.

Large-margin multiple kernel [formula omitted]-SVDD using Frank–Wolfe algorithm for novelty detection

Using a variable ℓp≥1-norm penalty on the slacks, the recently introduced ℓp-norm Support Vector Data Description (ℓp-SVDD) method has improved the performance in novelty detection over the baseline approach, sometimes remarkably. This work extends this modelling formalism in multiple aspects. First, a large-margin extension of the ℓp-SVDD method is formulated to enhance generalisation capability by maximising the margin between the positive and negative samples. Second, based on the Frank–Wolfe algorithm, an efficient yet effective method with predictable accuracy is presented to optimise the convex objective function in the proposed method. Finally, it is illustrated that the proposed approach can effectively benefit from a multiple kernel learning scheme to achieve state-of-the-art performance.The proposed method is theoretically analysed using Rademacher complexities to link its classification error probability to the margin and experimentally evaluated on several datasets to demonstrate its merits against existing methods.

Adaptive kernel density estimation weighted twin support vector machine and its sample screening method

SummaryIn twin support vector machines (TSVM), noise blurs the boundary between positive and negative samples, increasing the probability of classification errors. In this article, we propose an adaptive kernel density estimation weighted twin support vector machine(AKWTSVM). AKWTSVM uses KDE based on K‐nearest neighbor estimation to calculate the probability density of samples. It automatically selects the optimal bandwidth based on the local density of the samples to improve the robustness of the algorithm. However, TSVM has high time complexity, to reduce the time costs, a sample screening method is proposed for AKWTSVM, named AKWTSVM‐SSM, which is based on the overall distance and local density, and reduces the time costs of the algorithm by reducing the sample size while ensuring the accuracy of the algorithm. The experiment with differently scaled noise environments of 0%, 5%, 10%, 15%, and 20% on 12 UCI datasets validate the accuracy and running time of AKWTSVM and AKWTSVM‐SSM. Experimental results prove the effectiveness and robustness of AKWTSVM, the robustness of AKWTSVM‐SSM, and its applicability to large‐scale datasets.

Preoperative prediction of extrathyroidal extension: radiomics signature based on multimodal ultrasound to papillary thyroid carcinoma

BackgroundThere is a recognized need for additional approaches to improve the accuracy of extrathyroidal extension (ETE) diagnosis in papillary thyroid carcinoma (PTC) before surgery. Up to now, multimodal ultrasound has been widely applied in disease diagnosis. We investigated the value of radiomic features extracted from multimodal ultrasound in the preoperative prediction of ETE.MethodsWe retrospectively pathologically confirmed PTC lesions in 235 patients from January 2019 to April 2022 in our hospital, including 45 ETE lesions and 205 non-ETE lesions. MaZda software was employed to obtain radiomics parameters in multimodal sonography. The most valuable radiomics features were selected by the Fisher coefficient, mutual information, probability of classification error and average correlation coefficient methods (F + MI + PA) in combination with the least absolute shrinkage and selection operator (LASSO) method. Finally, the multimodal model was developed by incorporating the clinical records and radiomics features through fivefold cross-validation with a linear support vector machine algorithm. The predictive performance was evaluated by sensitivity, specificity, accuracy, F1 scores and the area under the receiver operating characteristic curve (AUC) in the training and test sets.ResultsA total of 5972 radiomics features were extracted from multimodal sonography, and the 13 most valuable radiomics features were selected from the training set using the F + MI + PA method combined with LASSO regression. The multimodal prediction model yielded AUCs of 0.911 (95% CI 0.866–0.957) and 0.716 (95% CI 0.522–0.910) in the cross-validation and test sets, respectively. The multimodal model and radiomics model showed good discrimination between ETE and non-ETE lesions.ConclusionRadiomics features based on multimodal ultrasonography could play a promising role in detecting ETE before surgery.

Latent Transition Cognitive Diagnosis Model With Covariates: A Three-Step Approach

To expand the use of cognitive diagnosis models (CDMs) to longitudinal assessments, this study proposes a bias-corrected three-step estimation approach for latent transition CDMs with covariates by integrating a general CDM and a latent transition model. The proposed method can be used to assess changes in attribute mastery status and attribute profiles and to evaluate the covariate effects on both the initial state and transition probabilities over time using latent (multinomial) logistic regression. Because stepwise approaches generally yield biased estimates, correction for classification error probabilities is considered in this study. The results of the simulation study showed that the proposed method yielded more accurate parameter estimates than the uncorrected approach. The use of the proposed method is also illustrated using a set of real data.

Application of mammography-based radiomics signature for preoperative prediction of triple-negative breast cancer

ObjectiveThis study is aimed to explore the value of mammography-based radiomics signature for preoperative prediction of triple-negative breast cancer (TNBC).Materials and methodsInitially, the clinical and X-ray data of patients (n = 319, age of 54 ± 14) with breast cancer (triple-negative—65, non-triple-negative—254) from the First Affiliated Hospital of Soochow University (n = 211, as a training set) and Suzhou Municipal Hospital (n = 108, as a verification set) from January 2018 to February 2021 are retrospectively analyzed. Comparing the mediolateral oblique (MLO) and cranial cauda (CC) mammography images, the mammography images with larger lesion areas are selected, and the image segmentation and radiomics feature extraction are then performed by the MaZda software. Further, the Fisher coefficients (Fisher), classification error probability combined average correlation coefficients (POE + ACC), and mutual information (MI) are used to select three sets of feature subsets. Moreover, the score of each patient’s radiomics signature (Radscore) is calculated. Finally, the receiver operating characteristic curve (ROC) is analyzed to calculate the AUC, accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of TNBC.ResultsA significant difference in the mammography manifestation between the triple-negative and the non-triple-negative groups (P < 0.001) is observed. The (POE + ACC)-NDA method showed the highest accuracy of 88.39%. The Radscore of triple-negative and non-triple-negative groups in the training set includes − 0.678 (− 1.292, 0.088) and − 2.536 (− 3.496, − 1.324), respectively, with a statistically significant difference (Z = − 6.314, P < 0.001). In contrast, the Radscore in the validation set includes − 0.750 (− 1.332, − 0.054) and − 2.223 (− 2.963, − 1.256), with a statistically significant difference (Z = − 4.669, P < 0.001). In the training set, the AUC, accuracy, sensitivity, specificity, positive predictive value and negative predictive value of TNBC include 0.821 (95% confidence interval 0.752–0.890), 74.4%, 82.5%, 72.5%, 41.2%, and 94.6%, respectively. In the validation set, the AUC, accuracy, sensitivity, specificity, positive predictive value and negative predictive value of TNBC are of 0.809 (95% confidence interval 0.711–0.907), 80.6%, 72.0%, 80.7%, 55.5%, and 93.1%, respectively.ConclusionIn summary, we firmly believe that this mammography-based radiomics signature could be useful in the preoperative prediction of TNBC due to its high value.

Soft-margin classification of object manifolds.

A neural population responding to multiple appearances of a single object defines a manifold in the neural response space. The ability to classify such manifolds is of interest, as object recognition and other computational tasks require a response that is insensitive to variability within a manifold. Linear classification of object manifolds was previously studied for max-margin classifiers. Soft-margin classifiers are a larger class of algorithms and provide an additional regularization parameter used in applications to optimize performance outside the training set by balancing between making fewer training errors and learning more robust classifiers. Here we develop a mean-field theory describing the behavior of soft-margin classifiers applied to object manifolds. Analyzing manifolds with increasing complexity, from points through spheres to general manifolds, a mean-field theory describes the expected value of the linear classifier's norm, as well as the distribution of fields and slack variables. By analyzing the robustness of the learned classification to noise, we can predict the probability of classification errors and their dependence on regularization, demonstrating a finite optimal choice. The theory describes a previously unknown phase transition, corresponding to the disappearance of a nontrivial solution, thus providing a soft version of the well-known classification capacity of max-margin classifiers. Furthermore, for high-dimensional manifolds of any shape, the theory prescribes how to define manifold radius and dimension, two measurable geometric quantities that capture the aspects of manifold shape relevant to soft classification.

A Fusion Method for Multisource Land Cover Products Based on Superpixels and Statistical Extraction for Enhancing Resolution and Improving Accuracy

The discrepancies in existing land cover data are relatively high, indicating low local precision and application limitations. Multisource data fusion is an effective way to solve this problem; however, the fusion procedure often requires resampling to unify the spatial resolution, causing a lower spatial resolution. To solve this problem, this study proposes a multisource product fusion mapping method of filtering training samples and product correction at a fine resolution. Based on the Superpixel algorithm, principal component analysis (PCA), and statistical extraction techniques, combined with the Google Earth Engine (GEE) platform, reliable land cover data were acquired. GEE and machine-learning algorithms correct the unreliable information of multiple products into a new land cover fusion result. Compared to the common method of extracting consistent pixels from existing products, our proposed method effectively removes nearly 38.75% of them, with a high probability of classification error. The overall accuracy of fusion in this study reached 85.80%, and the kappa coefficient reached 0.82, with an overall accuracy improvement of 11.75–24.17% and a kappa coefficient improvement of 0.16 to 0.3 compared to other products. For existing single-category products, we corrected the phenomenon of overinterpretation in inconsistent areas; the overall accuracy improvement ranged from 2.99% to 20.71%, while the kappa coefficient improvement ranged from 0.22 to 0.56. Thus, our proposed method can combine information from multiple products and serve as an effective method for large areas and even as a global land cover fusion product.

Statistical Hypothesis Testing Based on Machine Learning: Large Deviations Analysis

We study the performance -- and specifically the rate at which the error probability converges to zero -- of Machine Learning (ML) classification techniques. Leveraging the theory of large deviations, we provide the mathematical conditions for a ML classifier to exhibit error probabilities that vanish exponentially, say $\sim \exp\left(-n\,I + o(n) \right)$, where $n$ is the number of informative observations available for testing (or another relevant parameter, such as the size of the target in an image) and $I$ is the error rate. Such conditions depend on the Fenchel-Legendre transform of the cumulant-generating function of the Data-Driven Decision Function (D3F, i.e., what is thresholded before the final binary decision is made) learned in the training phase. As such, the D3F and, consequently, the related error rate $I$, depend on the given training set, which is assumed of finite size. Interestingly, these conditions can be verified and tested numerically exploiting the available dataset, or a synthetic dataset, generated according to the available information on the underlying statistical model. In other words, the classification error probability convergence to zero and its rate can be computed on a portion of the dataset available for training. Coherently with the large deviations theory, we can also establish the convergence, for $n$ large enough, of the normalized D3F statistic to a Gaussian distribution. This property is exploited to set a desired asymptotic false alarm probability, which empirically turns out to be accurate even for quite realistic values of $n$. Furthermore, approximate error probability curves $\sim \zeta_n \exp\left(-n\,I \right)$ are provided, thanks to the refined asymptotic derivation (often referred to as exact asymptotics), where $\zeta_n$ represents the most representative sub-exponential terms of the error probabilities.

Improved window adaptive gray level co-occurrence matrix for extraction and analysis of texture characteristics of pulmonary nodules

Background and ObjectiveIdentifying benign and malignant pulmonary nodules is essential for the early diagnosis of lung cancer and targeted surgical resection. This study aimed to differentiate benign from malignant pulmonary nodules based on computed tomography (CT) plain scan texture analysis technique. MethodsA total of 47 pulmonary nodules use the improved window adaptive gray level co-occurrence matrix (GLCM) algorithm to extract the texture characteristics of the area of interest. The Fisher coefficient (Fisher), classification error probability joint average correlation coefficient (POE+ACC), mutual information (MI), and the combination of above three methods joint (FPM) were used to select the best texture parameters set. After that, the analysis of the screened texture parameters was adopted. The B11 module provides four analytical methods, including raw data analysis (RDA), principal component analysis (PCA), linear discriminant analysis (LDA), and nonlinear discriminant analysis (NDA). The results were expressed in the form of misclassification rate (MCR). Region of curve (ROC) analysis was also performed on the selected optimal texture parameters. ResultsThe MCR of all the three texture feature extraction methods, Fisher, POE+ACC, and MI, were lower in differentiating benign from malignant pulmonary nodules. FPM method could further reduce the MCR. The NDA analysis had the lowest MCR for both of these three feature extraction methods. The MCR can be further reduced to 2.13% by the combination of NDA and FPM. The ROC curve showed that Perc.01% parameter had the highest AUC value and the most discriminative efficacy. ConclusionThe lowest MCR values were calculated by the FPM dimensionality reduction method and the NDA analysis method. The improved GLCM algorithm has a discriminative role in CT texture analysis of benign and malignant pulmonary nodules.

Radiomics Score Combined with ACR TI-RADS in Discriminating Benign and Malignant Thyroid Nodules Based on Ultrasound Images: A Retrospective Study.

This study aimed to explore the ability of combination model of ultrasound radiomics score (Rad-score) and the thyroid imaging, reporting and data system by the American College of Radiology (ACR TI-RADS) in predicting benign and malignant thyroid nodules (TNs). Up to 286 radiomics features were extracted from ultrasound images of TNs. By using the lowest probability of classification error and average correlation coefficients (POE + ACC) and the least absolute shrinkage and selection operator (LASSO), we finally selected four features to establish Rad-score (Vertl-RLNonUni, Vertl-GLevNonU, WavEnLH-s4 and WavEnHL-s5). DeLong’s test and decision curve analysis (DCA) showed that the method of combining Rad-score and ACR TI-RADS had the best performance (the area under the receiver operating characteristic curve (AUC = 0.913 (95% confidence interval (CI), 0.881–0.939) and 0.899 (95%CI, 0.840–0.942) in the training group and verification group, respectively), followed by ACR TI-RADS (AUC = 0.898 (95%CI, 0.863–0.926) and 0.870 (95%CI, 0.806–0.919) in the training group and verification group, respectively), and followed by Rad-score (AUC = 0.750 (95%CI, 0.704–0.792) and 0.750 (95%CI, 0.672–0.817) in the training group and verification group, respectively). We concluded that the ability of ultrasound Rad-score to distinguish benign and malignant TNs was not as good as that of ACR TI-RADS, and the ability of the combination model of Rad-score and ACR TI-RADS to discriminate benign and malignant TNs was better than ACR TI-RADS or Rad-score alone. Ultrasound Rad-score might play a potential role in improving the differentiation of malignant TNs from benign TNs.

Predicting response to non-small cell lung cancer immunotherapy using pre-treatment contrast-enhanced CT texture-based classification

Objective: To explore the value of pre-treatment contrast-enhanced computed tomography (CT)-based texture analysis in predicting response to non-small cell lung cancer (NSCLC) immunotherapy. Methods: From January to July 2018, a total of 51 lesions from 42 patients with advanced non-small cell lung cancer receiving immunotherapy at Shanghai Chest Hospital were selected in this retrospective study. Pre-treatment contrast-enhanced CT-based texture features were extracted by MaZda software. Ten optimal texture features were chosen based on three different methods: Fisher coefficient, mutual information measure (MI) and minimization of classification error probability combined average correlation coefficients(POE+ ACC), respectively. According to the efficacy of the first immunotherapy, 51 lesions were divided into non-progressive disease (non-PD, n=26) and progressive disease (PD, n=25). The differences were tested in each texture feature set between the two groups. The immunotherapy effects of target lesions were analyzed by principal component analysis(PCA), linear discriminant analysis (LDA) and nonlinear discriminant analysis (NDA). The sensitivity, specificity, accuracy, positive-predictive value (PPV) and negative-predictive value (NPV) were calculated. The area under the curve (AUC) was used to quantify the predictive accuracy of the three analysis models for each texture feature set and compare them with the actual classification results. Results: In all of three texture feature sets, the texture parameter differences of Perc.50%, Perc.90%, "S(5, 5)SumEntrp" and "S(4, 4)SumEntrp" were higher in PD group than those in non-PD group (all P<0.05). The classification result of texture feature set chosen by POE+ ACC and analyzed by NDA was identified as the best model (AUC=0.802, 95%CI: 0.674-0.930), and its sensitivity, specificity, accuracy, PPV and NPV were 72%, 88.5%, 80.4%, 85.7%, 76.7%, respectively. Conclusion: Pre-treatment contrast-enhanced CT-based texture characteristics of NSCLC may function as non-invasive biomarkers for the evaluation of response to immunotherapy.

Texture analysis based on U-Net neural network for intracranial hemorrhage identification predicts early enlargement

Background and objectiveEarly hemorrhage enlargement in hypertensive cerebral hemorrhage indicates a poor prognosis. This study aims to predict the early enlargement of cerebral hemorrhage through the intelligent texture analysis of cerebral hemorrhage after segmentation. MethodsA total of 54 patients with hypertensive intracerebral hemorrhage were selected and divided into enlarged hematoma (enlarged group) and non-enlarged hematoma (negative group). The U-Net Neural network model and contour recognition were used to extract the brain parenchymal region, and Mazda texture analysis software was used to extract regional features. The texture features were reduced by Fisher coefficient (Fisher), classification error probability combined average correlation coefficients (POE + ACC), and mutual information (MI) to select the best feature parameters. B11 module was used to analyze the selected features. The misclassified rate of feature parameters screened by different dimensionality reduction methods was calculated. ResultsThe neural network based on U-Net can accurately identify the lesion of cerebral hemorrhage. Among the 54 patients, 18 were in the enlarged group and 36 in the negative group. The parameters of gray level co-occurrence matrix and gray level run length matrix can be used to predict the enlargement of intracerebral hemorrhage. Among the features screened by Fisher, POE + ACC and MI, the texture features of MI showed the lowest misclassified rate, which was 0. ConclusionThe texture analysis based on U-Net neural network is helpful to predict the early expansion of hypertensive cerebral hemorrhage, and the parameters of gray level co-occurrence matrix and gray level run length matrix under MI dimensionality reduction have the most excellent predictive value.

Simulation of Random Differential Periodontitis Outcome Misclassification with Perfect Specificity

Misclassification of clinical periodontitis can occur by partial-mouth protocols, particularly when tooth-based case definitions are applied. In these cases, the true prevalence of periodontal disease is underestimated, but specificity is perfect. In association studies of periodontal disease etiology, misclassification by this mechanism is independent of exposure status (i.e., nondifferential). Despite nondifferential mechanisms, differential misclassification may be realized by virtue of random errors. To gauge the amount of uncertainty around the expectation of differential periodontitis outcome misclassification due to random error only, we estimated the probability of differential outcome misclassification, its magnitude, and expected impacts via simulation methods using values from the periodontitis literature. We simulated data sets with a binary exposure and outcome that varied according to sample size (200, 1,000, 5,000, 10,000), exposure effect (risk ratio; 1.5, 2), exposure prevalence (0.1, 0.3), outcome incidence (0.1, 0.4), and outcome sensitivity (0.6, 0.8). Using a Bernoulli trial, we introduced misclassification by randomly sampling individuals with the outcome in each exposure group and repeated each scenario 10,000 times. The probability of differential misclassification decreased as the simulation parameter values increased and occurred at least 37% of the time across the 10,000 repetitions. Across all scenarios, the risk ratio was biased, on average, toward the null when the sensitivity was higher among the unexposed and away from the null when it was higher among the exposed. The extent of bias for absolute sensitivity differences ≥0.04 ranged from 0.05 to 0.19 regardless of simulation parameters. However, similar trends were not observed for the odds ratio where the extent and direction of bias were dependent on the outcome incidence, sensitivity of classification, and effect size. The results of this simulation provide helpful quantitative information to guide interpretation of findings in which nondifferential outcome misclassification mechanisms are known to be operational with perfect specificity. Measurement of periodontitis can suffer from classification errors, such as when partial-mouth protocols are applied. In this case, specificity is perfect and sensitivity is expected to be nondifferential, leading to an expectation for no bias when studying periodontitis etiologies. Despite expectation, differential misclassification could occur from sources of random error, the effects of which are unknown. Proper scrutiny of research findings can occur when the probability and impact of random classification errors are known.

Pre-Immunotherapy Contrast-Enhanced CT Texture-Based Classification: A Useful Approach to Non-Small Cell Lung Cancer Immunotherapy Efficacy Prediction.

Objective: To investigate the utility of the pre-immunotherapy contrast-enhanced CT-based texture classification in predicting response to non-small cell lung cancer (NSCLC) immunotherapy treatment.Methods: Sixty-three patients with 72 lesions who received immunotherapy were enrolled in this study. We extracted textures including histogram, absolute gradient, run-length matrix, gray-level co-occurrence matrix, autoregressive model, and wavelet transform from pre-immunotherapy contrast-enhanced CT by using Mazda software. Three different methods, namely, Fisher coefficient, mutual information measure (MI), and minimization of classification error probability combined average correlation coefficients (POE + ACC), were performed to select 10 optimal texture feature sets, respectively. The patients were divided into non-progressive disease (non-PD) and progressive disease (PD) groups. t-test or Mann–Whitney U-test was performed to test the differences in each texture feature set between the above two groups. Each texture feature set was analyzed by principal component analysis (PCA), linear discriminant analysis (LDA), and non-linear discriminant analysis (NDA). The area under the curve (AUC) was used to quantify the predictive accuracy of the above three analysis models for each texture feature set, and the sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) were also calculated, respectively.Results: Among the three texture feature sets, the texture parameter differences of kurtosis (2.12 ± 3.92 vs. 0.78 ± 1.10, p = 0.047), “S(2,2)SumEntrp” (1.14 ± 0.31 vs. 1.24 ± 0.12, p = 0.036), and “S(1,0)SumEntrp” (1.18 ± 0.27 vs. 1.28 ± 0.11, p = 0.046) between the non-PD and PD group were statistically significant (all p < 0.05). The classification result of texture feature set selected by POE + ACC and analyzed by NDA was identified as the best model (AUC = 0.812, 95% CI: 0.706–0.919) with a sensitivity, specificity, accuracy, PPV, and NPV of 88.2, 76.3, 81.9, 76.9, and 87.9%, respectively.Conclusion: Pre-immunotherapy contrast-enhanced CT-based texture provides a new method for clinical evaluation of the NSCLC immunotherapy efficacy prediction.

Re-Weighting Large Margin Label Distribution Learning for Classification.

Label ambiguity has attracted quite some attention among the machine learning community. The latterly proposed Label Distribution Learning (LDL) can handle label ambiguity and has found wide applications in real classification problems. In the training phase, an LDL model is learned first. In the test phase, the top label(s) in the label distribution predicted by the learned LDL model is (are) then regarded as the predicted label(s). That is, LDL considers the whole label distribution in the training phase, but only the top label(s) in the test phase, which likely leads to objective inconsistency. To avoid such inconsistency, we propose a new LDL method Re-Weighting Large Margin Label Distribution Learning (RWLM-LDL). First, we prove that the expected L1-norm loss of LDL bounds the classification error probability, and thus apply L1-norm loss as the learning metric. Second, re-weighting schemes are put forward to alleviate the inconsistency. Third, large margin is introduced to further solve the inconsistency. The theoretical results are presented to showcase the generalization and discrimination of RWLM-LDL. Finally, experimental results show the statistically superior performance of RWLM-LDL against other comparing methods.

Computed tomography in the diagnosis of intraperitoneal effusions: The role of texture analysis.

The morphological changes suggesting peritoneal carcinomatosis are inconsistent and may be visible only at later stages of the disease. However, malignant ascites represents an early sign, and this fluid exhibits specific histological characteristics. This study aimed to quantify the fluid properties on computed tomography (CT) images of intraperitoneal effusions through texture analysis and evaluate its utility in differentiating benign from malignant collections. Fifty-two patients with histologically proven benign (n=29) and malignant (n=23) intraperitoneal effusions who underwent CT examinations were retrospectively included. Texture analysis of the fluid component was performed on the non-enhanced phase of each examination using dedicated software. Fisher and the probability of classification error and average correlation coefficients were used to select two sets of ten texture features, whose ability to distinguish between the two types of collections were tested using a k-nearest-neighbor classifier. Also, each of the selected feature’s diagnostic power was assessed using univariate and receiver operating characteristics analysis based on the calculation of the area under the curve. The k-nearest-neighbor classifier was able to distinguish between the two entities with 71.15% accuracy, 73.91% sensitivity, and 68.97% specificity. The highest-ranked texture parameter was Inverse Difference Moment (p=0.0023; area under the curve=0.748), based on which malignant collections could be diagnosed with 95.65% sensitivity and 44.83% specificity. Although successful, the texture assessment of benign and malignant collections is less effective in reflecting the cytological differences between the two groups.