Performance Of Different Prediction Models Research Articles

An empirically based object-oriented testing using Machine learning

INTRODUCTION: The rapid growth of machine learning has the potential to revolutionize various industries and applications by automating complex tasks and enhancing efficiency. Effective software testing is crucial for ensuring software quality and minimizing resource expenses in software engineering. Machine learning techniques play a vital role in software testing by aiding in test case prioritization, predicting software defects, and analyzing test results. OBJECTIVES: The primary objective of this study is to explore the use of machine learning algorithms for software defect prediction. METHODS: Machine Learning models including Random Forest Classifier, Logistic Regression, K Nearest Neighbors, Gradient Boosting Classifiers, Catboost Classifier, and Convolutional Neural Networks have been employed for the study. The dataset includes a wide range of features relevant to software defect prediction and evaluates the performance of different prediction models. The study also focussed on developing hybrid models using stacking classifiers, which combine multiple individual models to improve accuracy. RESULTS: The experimental results show that the hybrid models combining CatBoost and Convolutional Neural Network have outperformed individual models, achieving the highest accuracy of 89.5%, highlighting the effectiveness of combining machine learning algorithms for software defect prediction. CONCLUSION: In conclusion, this study sheds light on the pivotal role of machine learning in enhancing software defect prediction.

Deformation extent prediction of roadway roof during non-support period using support vector regression combined with swarm intelligent bionic optimization algorithms

Settlement deformation of the roof during non-support period in deep roadway excavation poses the most immediate hazard to coal mine roadway engineering. Thus, accurately predicting the roof’s displacement during this period is crucial in minimizing potential risks. Considering the intricate nature of factors that influence deep roadway engineering, this research employs Support Vector Regression (SVR) as the primary tool, coupled with four swarm intelligent bionic optimization algorithms: Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO) and Grasshopper Optimization Algorithm (GOA) to enhance the prediction performance and optimize hyper-parameters. The prediction models involve seven influential factors that are selected from the underground test roadway. These factors form a comprehensive evaluation system which is based on artificial intelligence techniques. To evaluate the performance of different prediction models, three error correlation coefficient indices are used, include root mean square error (RMSE), variance accounted for (VAF) and coefficient of determination (R2). Of all the models, the GWO-e-SVR model exhibits the best overall performance in predicting deformation extent of the roof during non-support period. With a swarm size of 100, the RMSE, VAF and R2 of training datasets are 0.3126, 15.05% and 0.8767, respectively. The RMSE, VAF and R2 of testing datasets are 0.3245, 16.43% and 0.8655, respectively. The model demonstrated outstanding predictive accuracy in small-scale experiments, signifying its strong generalization and robustness. Subsequently, a sensitivity analysis was conducted to expound on the impact of conclusive input parameters on the end target output, specifically the duration of non-support in the space–time effect concerning roof’s settlement during the non-supported state. These findings hold significant implications for the upcoming support construction scheme design and reinforcement support operation.

A Model Combining Conventional Ultrasound Characteristics, Strain Elastography and Clinicopathological Features to Predict Ki-67 Expression in Small Breast Cancer.

To establish a predictive model incorporating conventional ultrasound, strain elastography and clinicopathological features for Ki-67 expression in small breast cancer (SBC) which defined as maximum diameter less than2 cm. In this retrospective study, 165 SBC patients from our hospital were allocated to a high Ki-67 group (n = 104) and a low Ki-67 group (n = 61). Multivariate regression analysis was performed to identify independent indicators for developing predictive models. The area under the receiver operating characteristic (AUC) curve was also determined to establish the diagnostic performance of different predictive models. The corresponding sensitivities and specificities of different models at the cutoff value were compared. Conventional ultrasound parameters (spiculated margin, absence of posterior shadowing and Adler grade 2-3), strain elastic scores and clinicopathological information (HER2 positive) were significantly correlated with high expression of Ki-67 in SBC (all p < .05). Model 2, which incorporated conventional ultrasound features and strain elastic scores, yielded good diagnostic performance (AUC = 0.774) with better sensitivity than model 1, which only incorporated ultrasound characteristics (78.85%vs. 55.77%, p = .000), with specificities of 77.05% and 62.30% (p = .035), respectively. Model 3, which incorporated conventional ultrasound, strain elastography and clinicopathological features, yielded better performance (AUC = 0.853) than model 1 (AUC = 0.694) and model 2 (AUC = 0.774), and the specificity was higher than model 1 (86.89% vs. 77.05%, p = .001). The predictive model combining conventional ultrasound, strain elastic scores and clinicopathological features could improve the predictive performance of Ki-67 expression in SBC.

Analysis and prediction of railway track longitudinal level using multiple machine learning methods

With the fast development of computer science, many prediction models based on machine learning methods have been used in the railway industry, which can better predict the random characteristics in track degradation, plan maintenance activities, and eventually meet the requirement of railway transportation. However, the applicability and generality of these models are unclear and the comparative analysis of these models on the random railway track is rare, which makes it hard for railway engineers to choose the most suitable prediction models in practice. In this paper, the track longitudinal level of a section of a railway measured monthly by the rail infrastructure alignment acquisition system for 1.5 years in the Netherlands has been analysed using multiple mathematical methods. After that, three machine learning-based prediction models were developed to predict the future development of the track longitudinal level, using support vector machine, grey model and deep neural network. The prediction performance of different prediction models is compared and discussed. Recommendations for choosing prediction models and further development are provided.

Spatial prediction of soil sand content at various sampling density based on geostatistical and machine learning algorithms in plain areas

Accurate prediction of the spatial distribution of soil sand content is a pre-requisite for land use management, soil quality evaluation and erosion control, as it determines the transport and movement of soil water, fertilizer, air and heat. Digital soil mapping (DSM) is extensively employed for predicting soil properties. However, practical research is required to address the challenge of selecting an optimal prediction model that is both cost-effective and accurate at a specific sampling density. In this study, topsoil samples were collected from 2,848 sampling points in the eastern plains of China (107,200 km2). The performance of different prediction models for mapping soil sand content was compared at 12 levels of sampling density. Moreover, the geographical detector, a statistical method used to assess the spatial stratified heterogeneity of variables, was adopted to determine the major drivers of spatial variation in soil sand content. The results indicated that climate factors are the major drivers of the spatial variability in soil sand content. For the 100% sample size (26.57 samples/103 km2), the geostatistical models that did not depend on environmental variables (ordinary kriging, sequential Gaussian simulation) performed best, followed by the machine learning models (random forest, cubist and support vector machine) and the geostatistical model with environmental variables (co-kriging). Sampling density had a considerable impact on model accuracy, and the advantages of machine learning models became apparent when sampling densities were below 20% (5.31 samples/103 km2). Therefore, the best combination of prediction model and sampling density should be selected to obtain maps of soil sand content economically and accurately. This study provides a valuable reference for the selection of prediction methods in the practical application of DSM.

Intelligent flood forecasting and warning: a survey

Accurately predicting the magnitude and timing of floods is an extremely challenging problem for watershed management, as it aims to provide early warning and save lives. Artificial intelligence for forecasting has become an emerging research field over the past two decades, as computer technology and related areas have been developed in depth. In this paper, three typical machine learning algorithms for flood forecasting are reviewed: supervised learning, unsupervised learning, and semi-supervised learning. Special attention is given to deep learning approaches due to their better performance in various prediction tasks. Deep learning networks can represent flood behavior as powerful and beneficial tools. In addition, a detailed comparison and analysis of the multidimensional performance of different prediction models for flood prediction are presented. Deep learning has extensively promoted the development of real-time accurate flood forecasting techniques for early warning systems. Furthermore, the paper discusses the current challenges and future prospects for intelligent flood forecasting.

Analysis of Total Soil Nutrient Content with X-ray Fluorescence Spectroscopy (XRF): Assessing Different Predictive Modeling Strategies and Auxiliary Variables

The difference in the matrix present in soil samples from different areas limits the performance of nutrient analysis via XRF sensors, and only a few strategies to mitigate this effect to ensure an accurate analysis have been proposed so far. In this context, this research aimed to compare the performance of different predictive models, including a simple linear regression (RS), multiple linear regression (MLR), partial least-squares regression (PLS), and random forest (RF) models for the prediction of Ca and K in agricultural soils. RS models were evaluated on XRF data without (RS1) and with (RS2) Compton normalization. In addition, it was assessed whether using soil texture information and/or vis–NIR spectra as auxiliary variables would optimize the predictive performance of the models. The results showed that all strategies allowed the mitigation of the matrix effect to some degree, enabling the determination of their Ca and K contents with excellent predictive performance (R2 ≥ 0.84). The best performance was obtained using RS2 for the Ca prediction (R2 = 0.92, RSME = 48.25 mg kg−1 and relative improvement (RI) of 52.3% compared to RS1) and using an RF for the K prediction (R2 = 0.84, RSME = 17.43 mg kg−1 and RI of 24.3% compared to RS1). The results indicated that sophisticated models did not always perform better than linear models. Furthermore, using texture data and vis–NIR spectra as auxiliary data was promising only for the K prediction, which showed an error reduction in the order of 10%, contrasting with the Ca prediction, which did not reduce the prediction error by more than 1%. The best modeling approach in our study proved to be attribute-specific. These results give further insight into the development of intelligence modeling strategies for sensor-based soil analysis.

Predicting Emotion with Biosignals: A Comparison of Classification and Regression Models for Estimating Valence and Arousal Level Using Wearable Sensors.

This study aims to predict emotions using biosignals collected via wrist-worn sensor and evaluate the performance of different prediction models. Two dimensions of emotions were considered: valence and arousal. The data collected by the sensor were used in conjunction with target values obtained from questionnaires. A variety of classification and regression models were compared, including Long Short-Term Memory (LSTM) models. Additionally, the effects of different normalization methods and the impact of using different sensors were studied, and the way in which the results differed between the study subjects was analyzed. The results revealed that regression models generally performed better than classification models, with LSTM regression models achieving the best results. The normalization method called baseline reduction was found to be the most effective, and when used with an LSTM-based regression model it achieved high accuracy in detecting valence (mean square error = 0.43 and R2-score = 0.71) and arousal (mean square error = 0.59 and R2-score = 0.81). Moreover, it was found that even if all biosignals were not used in the training phase, reliable models could be obtained; in fact, for certain study subjects the best results were obtained using only a few of the sensors.

A nomogram based on combining clinical features and contrast enhanced ultrasound is not able to identify Her-2 over-expressing cancer from other breast cancers.

The aim of this study was to evaluate whether a predictive model based on a contrast enhanced ultrasound (CEUS)-based nomogram and clinical features (Clin) could differentiate Her-2-overexpressing breast cancers from other breast cancers. A total of 152 pathology-proven breast cancers including 55 Her-2-overexpressing cancers and 97 other cancers from two units that underwent preoperative CEUS examination, were included and divided into training (n = 102) and validation cohorts (n = 50). Multivariate regression analysis was utilized to identify independent indicators for developing predictive nomogram models. The area under the receiver operating characteristic (AUC) curve was also calculated to establish the diagnostic performance of different predictive models. The corresponding sensitivities and specificities of different models at the cutoff nomogram value were compared. In the training cohort, 7 clinical features (menstruation, larger tumor size, higher CA153 level, BMI, diastolic pressure, heart rate and outer upper quarter (OUQ)) + enlargement in CEUS with P < 0.2 according to the univariate analysis were submitted to the multivariate analysis. By incorporating clinical information and enlargement on the CEUS pattern, independently significant indicators for Her-2-overexpression were used for further predictive modeling as follows: Model I, nomogram model based on clinical features (Clin); Model II, nomogram model combining enlargement (Clin + Enlargement); Model III, nomogram model based on typical clinical features combining enlargement (MC + BMI + diastolic pressure (DP) + outer upper quarter (OUQ) + Enlargement). Model II achieved an AUC value of 0.776 at nomogram cutoff score value of 190, which was higher than that of the other models in the training cohort without significant differences (all P>0.05). In the test cohort, the diagnostic efficiency of predictive model was poor (all AUC<0.6). In addition, the sensitivity and specificity were not significantly different between Models I and II (all P>0.05), in either the training or the test cohort. In addition, Clin exhibited an AUC similar to that of model III (P=0.12). Moreover, model III exhibited a higher sensitivity (70.0%) than the other models with similar AUC and specificity, only in the test cohort. The main finding of the study was that the predictive model based on a CEUS-based nomogram and clinical features could not differentiate Her-2-overexpressing breast cancers from other breast cancers.

A Novel Machine Learning Framework for Analyzing Performance of Different Prediction Models by Using Automatic Malware Detection (AMD) Algorithm

The efficiency of the machine learning techniques in context to the detection of malware's has been proved by state-of-the-art research works. Since adversaries are developing various kinds of malware, its detection has become a challenging task. Unfortunately, there is every growing threat to cyberspace due to different kinds of malicious programs termed as malware. Malware detection has become challenging task now days. Supervised Machine Learning Algorithms are used to detect malware in emerging technologies such as Internet of Things (IoT) and distributed computing, there is seamless integration of heterogeneous applications with interoperability The objective of this paper is to explore and evaluate different ML models with empirical study. In this paper, we proposed a ML framework for analyzing performance of different prediction models. An algorithm known as Machine Learning based Automatic Malware Detection (ML-AMD) is proposed. This algorithm is used to realize the framework with supervised learning. This empirical study has resulted in knowledge about ML models such as Decision Tree (DT), Logistic Regression (LR), Random Forest (RF), Multilayer Perception (MLP) and Gradient Boosting (GB). Random Forest model has exhibited highest accuracy with 97.96%. The research outcomes in this paper help in triggering further investigations towards automatic detection of malware. Keywords: malware detection, machine learning, decision tree, logistic regression, random forest, multilayer perception, gradient boosting

Warehouse equipment maintenance strategy based on the prediction of total amount of stock in & out operations

Warehouse is an important part of the supply chain. However, the frequent failure of warehouse equipment greatly restricts scheduling efficiency. To achieve efficient operation and reduce the frequency of sudden failures, a new strategy for warehouse center maintenance based on the prediction of total amount of stock in & out operations is proposed. The feasibility of this strategy has been proved by using the historical data of operation and maintenance for three auto parts warehouse centers in South China, North China and East China. The orthogonal experimental method, which can select the preferred combination of hyperparameters quickly, is used to implement hyperparameter estimation for the predictive models. Based on the performance of different predictive models on the warehouse operation datasets, we determined the best model to predict the future total amount of warehouse stock in & out operations. The results show that compared with other models, the bidirectional long-short-term memory (Bi-LSTM) model obtains the highest accuracy on the test set, which is used for subsequent prediction. According to the prediction results, there is no need to arrange an overhaul plan for the North China and East China warehouse centers in the short term. However, the South China warehouse center may have frequent equipment failures in two months. It is necessary to formulate corresponding maintenance plans in advance.

MSR64 Utility Analysis Methods: Can One Method Fit It All or Is a Case-By-Case Approach Required?

Health-related quality-of-life data has distributional properties that raise numerous statistical challenges. Our objective was to compare the performance of different predictive models in handling health utility data.

The automation of the development of classification models and improvement of model quality using feature engineering techniques

Recently pipelines of machine learning-based classification models have become important to codify, orchestrate, and automate the workflow to produce an effective machine learning model. In this article, we propose a framework that combines feature engineering techniques such as data imputation, transformation, and class balancing to compare the performance of different prediction models and select the best final model based on predefined parameters. The proposed framework is extendable and configurable by adding algorithms supported by the CARET package implemented in the R programming language. This framework can generate different machine learning models, which provide comparable results compared to other studies. The framework allows practitioners and researchers to automatically generate different classification models. This research used High-Resolution Orbitrap-based Mass Spectrometers (HRMS) data to create automated prediction models for the first time in literature. We demonstrated the applicability of feature engineering techniques such as data imputation, transformation (e.g., scaling, centering, etc.), and data balancing using several case studies and the proposed semi-automated framework. We showed how the initial prediction models can be improved using the proposed framework.

Validation and comparison of multiple risk scores for prediction of symptomatic intracerebral hemorrhage after intravenous thrombolysis in VISTA.

Prediction models/scores may help to identify patients at high risk of symptomatic intracerebral hemorrhage (sICH) after intravenous thrombolysis. We aimed to validate and compare the performance of different prediction models for sICH after thrombolysis using direct model estimation in the Virtual International Stroke Trials Archive (VISTA). We searched PubMed for potentially eligible prediction models from inception to 1 June 2019. Simple and practical models/scores were validated in VISTA. The primary outcome was sICH based on two criteria (National Institute of Neurological Diseases and Stroke, NINDS; Safe Implementation of Thrombolysis in Stroke-Monitoring Study, SITS-MOST) and the secondary outcome was parenchymal hematoma (PH). The discrimination performance of each model was evaluated using area under the curve (AUC) and calibration was evaluated by Hosmer-Lemeshow goodness-of-fit tests. We found 13 prediction models and five models (HAT, MSS, SPAN-100, GRASPS and THRIVE) were finally validated in VISTA. A total of 1884 participants were eligible for our study, of whom the proportion with sICH was 4.6% (87/1884) per NINDS and 3.9% (73/1884) per SITS-MOST, and with PH was 11.3% (213/1884). MSS and GRASPS had the greatest predictive ability for sICH (NINDS criteria: MSS AUC 0.7, 95% CI 0.63-0.77, p < 0.001; GRASPS AUC 0.69, 95% CI 0.63-0.76, p < 0.001; SITS-MOST criteria: MSS, AUC 0.76, 95% CI 0.68-0.85, p < 0.001; GRASPS, AUC 0.79, 95% CI 0.71-0.87, p < 0.001). Similar results were found for PH (MSS AUC 0.68, 95% CI 0.64-0.73, p = 0.017; GRASPS AUC 0.68, 95% CI 0.63-0.72, p = 0.017). The calibration of each model was almost good. MSS and GRASPS had good discrimination and calibration for sICH and PH after thrombolysis as assessed in VISTA. These two models could be used in clinical practice and clinical trials to identity individuals with high risk of sICH.

Employing a Time Series Forecasting Model for Tourism Demand Using ANFIS

Forecasting the future trends is of utmost importance for managers and decision makers in different sectors. Scholars thus have introduced various techniques to the service industry aiming at employing a prediction model with ultimate accuracy and high efficiency. The literature proves that adaptive neuro-fuzzy inference systems (ANFIS) are the most efficiency models. However, the literature lacks reports on how ANFIS parameters may affect the accuracy of the system. Employing tourist arrival records to Cyprus between 2015 and 2019, this study has developed an ANFIS system to evaluate the accuracy performance of different prediction models with varied number of inputs and number or type of membership functions. Results show that the forecasting accuracy of a model with four inputs and four membership functions when the type of membership functions is Gaussian is relatively better than other models. In other words, it can be concluded that the forecast model with four inputs and four Gaussian membership functions is ultimate with the most accurate prediction record with reference to MAE, RMSE, and MAPE. The results of this study may be significant for senior managers and decision-makers of the tourism industry.

Inclusion of sarcopenia improves the prognostic value of MELD score in patients after transjugular intrahepatic portosystemic shunt.

ObjectiveTo explore the predictive value of model for end-stage liver disease (MELD)-Sarcopenia score for survival of cirrhotic patients after transjugular intrahepatic portosystemic shunt (TIPS) placement.Methods289 patients who underwent TIPS between February 2016 and December 2020 were included, they were divided into the sarcopenia group (n = 138) and non-sarcopenia group (n = 151) according to whether they were complicated with sarcopenia. Kaplan–Meier curve was used to analyze and compare the prognosis of the above two groups and multivariate Cox regression analysis was used to identify the independent prognostic factors. The performance of different predictive models was compared using C-index.ResultsDuring the follow-up, Kaplan–Meier analyses indicated that cumulative survival was significantly lower in sarcopenia group than that in non-sarcopenia group [74.6% vs. 92.7%, HR, 0.24 (95% confidence interval (CI), 0.12–0.46), Log-rank P < 0.001]. After multivariate Cox analysis, age [HR, 1.040 (95% CI, 1.015–1.065), P = 0.002], sarcopenia [HR, 3.948 (95% CI, 1.989–7.838), P < 0.001], albumin [HR, 0.945 (95% CI, 0.897–0.997), P = 0.037], and MELD score [HR, 1.156 (95% CI, 1.097–1.217), P < 0.001] were identified as the independent risk factors for mortality after TIPS. The C-indexes of MELD-Sarcopenia, Child-Pugh, MELD, MELD-Na, and the Freiburg index of post-TIPS survival (FIPS) scores were 0.782, 0.688, 0.719, 0.734, and 0.770, respectively.ConclusionSarcopenia is independently correlated with post-TIPS mortality, and MELD-Sarcopenia score showed the best performance in predicting post-TIPS mortality than the traditional predictive models.

Ultrasound high-definition microvasculature imaging with novel quantitative biomarkers improves breast cancer detection accuracy

ObjectivesTo overcome the limitations of power Doppler in imaging angiogenesis, we sought to develop and investigate new quantitative biomarkers of a contrast-free ultrasound microvasculature imaging technique for differentiation of benign from malignant pathologies of breast lesion.MethodsIn this prospective study, a new high-definition microvasculature imaging (HDMI) was tested on 521 patients with 527 ultrasound-identified suspicious breast masses indicated for biopsy. Four new morphological features of tumor microvessels, microvessel fractal dimension (mvFD), Murray’s deviation (MD), bifurcation angle (BA), and spatial vascularity pattern (SVP) as well as initial biomarkers were extracted and analyzed, and the results correlated with pathology. Multivariable logistic regression analysis was used to study the performance of different prediction models, initial biomarkers, new biomarkers, and combined new and initial biomarkers in differentiating benign from malignant lesions.ResultsThe new HDMI biomarkers, mvFD, BA, MD, and SVP, were statistically significantly different in malignant and benign lesions, regardless of tumor size. Sensitivity and specificity of the new biomarkers in lesions > 20 mm were 95.6% and 100%, respectively. Combining the new and initial biomarkers together showed an AUC, sensitivity, and specificity of 97% (95% CI: 95–98%), 93.8%, and 89.2%, respectively, for all lesions regardless of mass size. The classification was further improved by adding the Breast Imaging Reporting and Data System (BI-RADS) score to the prediction model, showing an AUC, sensitivity, and specificity of 97% (95% CI: 95–98%), 93.8%, and 89.2%, respectively.ConclusionThe addition of new quantitative HDMI biomarkers significantly improved the accuracy in breast lesion characterization when used as a complementary imaging tool to the conventional ultrasound.Key Points• Novel quantitative biomarkers extracted from tumor microvessel images increase the sensitivity and specificity in discriminating malignant from benign breast masses.• New HDMI biomarkers Murray’s deviation, bifurcation angles, microvessel fractal dimension, and spatial vascularity pattern outperformed the initial biomarkers.• The addition of BI-RADS scores based on US descriptors to the multivariable analysis using all biomarkers remarkably increased the sensitivity, specificity, and AUC in all size groups.

Prediction of blasting mean fragment size using support vector regression combined with five optimization algorithms

The main purpose of blasting operation is to produce desired and optimum mean size rock fragments. Smaller or fine fragments cause the loss of ore during loading and transportation, whereas large or coarser fragments need to be further processed, which enhances production cost. Therefore, accurate prediction of rock fragmentation is crucial in blasting operations. Mean fragment size (MFS) is a crucial index that measures the goodness of blasting designs. Over the past decades, various models have been proposed to evaluate and predict blasting fragmentation. Among these models, artificial intelligence (AI)-based models are becoming more popular due to their outstanding prediction results for multi-influential factors. In this study, support vector regression (SVR) techniques are adopted as the basic prediction tools, and five types of optimization algorithms, i.e. grid search (GS), grey wolf optimization (GWO), particle swarm optimization (PSO), genetic algorithm (GA) and salp swarm algorithm (SSA), are implemented to improve the prediction performance and optimize the hyper-parameters. The prediction model involves 19 influential factors that constitute a comprehensive blasting MFS evaluation system based on AI techniques. Among all the models, the GWO-v-SVR-based model shows the best comprehensive performance in predicting MFS in blasting operation. Three types of mathematical indices, i.e. mean square error (MSE), coefficient of determination (R2) and variance accounted for (VAF), are utilized for evaluating the performance of different prediction models. The R2, MSE and VAF values for the training set are 0.8355, 0.00138 and 80.98, respectively, whereas 0.8353, 0.00348 and 82.41, respectively for the testing set. Finally, sensitivity analysis is performed to understand the influence of input parameters on MFS. It shows that the most sensitive factor in blasting MFS is the uniaxial compressive strength.

Development of total suspended matter prediction in waters using fractional-order derivative spectra

More and more hyper-spectral satellites will be used to estimate total suspended matter (TSM) in waters instead of multi-spectral satellites, such as China's Gaofen-5 and Zhuhai-1. Although they have not been widely used because of the consistency of sampling and image time. Hence, the study based on measured hyper-spectroscopy is important for applying to hyper-spectral satellites. Fractional-order derivatives (FODs) considers more detailed spectral information, and it is a better spectral preprocessing method than conventional integer-order derivatives. The application and analysis of FODs for spectra in waters is rare. If FOD is successfully applied to estimate TSM, the TSM mapping with FOD using hyper-spectral satellites will be meaningful. Based on these points, this study aimed to apply FOD to predict TSM and to prove the prediction feasibility of FOD in waters. Different prediction models and eight FOD transformation processes with increment of 0.25 per step for 392 spectral reflectance data from China were used and compared. The prediction models include the optimum models of the single wavelength, ratio index, difference index and TSM index at each FOD order, and the random forest (RF) model with all wavelengths was also used. Discrete wavelet transform (DWT) was used to reduce noise and improve the model accuracy after using FOD. Our results achieved the followings First, FOD enhanced spectral characteristics at 500–600 nm and 800 nm that were affected by TSM. Second, the correlation between TSM and FOD spectra was enhanced (e.g., the correlation coefficients of 19 wavelengths (789–807 nm) of 0.75-order were higher than 0.8 but the original spectra were not). Third, FOD improved the performance of different prediction models, and the RF model from 0.5-order to 1.25-order derivative spectra all led good results (). Fourth, DWT can reduce the noise and improve the performance, and FOD-DWT model of 1.25-order led the R2 of 0.84, RMSE of 16.30 and MAPE of 78.62 in validation. Overall, our results suggest that FOD can improve the prediction performance for most models, and the optimum order of some models is not integer. Our results also provide a reference for predicting other water quality parameters and mapping these parameters using hyper-spectral satellites. The accurate estimation of TSM is helpful for protecting ecological and social environments.

A Nomogram Based on Combining Clinical Features and Contrast Enhanced Ultrasound LI-RADS Improves Prediction of Microvascular Invasion in Hepatocellular Carcinoma.

PurposesTo establish a predictive model incorporating clinical features and contrast enhanced ultrasound liver imaging and reporting and data system (CEUS LI-RADS) for estimation of microvascular invasion (MVI) in hepatocellular carcinoma (HCC) patients.MethodsIn the retrospective study, 127 HCC patients from two hospitals were allocated as training cohort (n=98) and test cohorts (n=29) based on cutoff time-point, June 2020. Multivariate regression analysis was performed to identify independent indicators for developing predictive nomogram models. The area under receiver operating characteristic (AUC) curve was also determined to establish the diagnostic performance of different predictive models. Corresponding sensitivities and specificities of different models at the cutoff nomogram value were compared.ResultsIn the training cohort, clinical information (larger tumor size, higher AFP level) and CEUS LR-M were significantly correlated with the presence of MVI (all p<0.05). By incorporating clinical information and CEUS LR-M, the predictive model (LR-M+Clin) achieved a desirable diagnostic performance (AUC=0.80 and 0.84) in both cohorts at nomogram cutoff score value of 89. The sensitivity of LR-M+Clin when predicting MVI in HCC patients was higher than that of the clinical model alone (86.7% vs. 46.7%, p=0.027), while specificities were 78.6% and 85.7% (p=0.06), respectively, in the test cohort. In addition, LR-M+Clin exhibited similar AUC and specificity, but a significantly higher sensitivity (86.7%) than those of LR-M alone and LR-5(No)+Clin (both sensitivities=73.3%, both p=0.048).ConclusionThe predictive model incorporating CEUS LR-M and clinical features was able to predict the MVI status of HCC and is a potential reliable preoperative tool for informing treatment.