Features Matrix Research Articles

Prognostic Impact of a Histologic Grading Scheme in Dogs Diagnosed With Rib Chondrosarcoma.

Data regarding the outcome of canine rib chondrosarcoma is sparse and varied. While grade of tumour is associated with outcome for canine appendicular chondrosarcoma, the association of grade with outcome for canine rib chondrosarcoma is unclear. This study aimed to correlate the grade of canine rib chondrosarcoma with median survival time. Retrospectively, cases of primary rib chondrosarcoma were identified, and tumours were graded based on a 3-tier adapted human grading scheme. Twenty-two patients were included in the survival analysis. The median survival time was 1427 days (range: 27-3354 days). This was not significantly different for patients with grade I versus II versus III (p = 0.82), grade I-II versus III (p = 0.34), or grade I versus II-III (p = 0.49). No variables assessed including age, weight, tumour location (cranial vs. caudal thorax; left vs. right hemithorax), tumour location on rib (proximal, middle, and distal), radiographic appearance (lytic, proliferative, or mixed), elevated serum alkaline phosphatase activity, grade, grade specific histologic features (matrix production, architecture, pleomorphism, cellularity, necrosis, and total score), adjunct therapy post-surgical excision, development of metastatic disease post-surgery, or local recurrence post-surgery were found to impact the risk of death due to chondrosarcoma. In this limited group of patients, the grading scheme reported here, and the other variables assessed did not appear to offer additional prognostic information. However, this data must be interpreted considering the small sample size and thus low statistical power. Additional studies are needed to determine the true impact of grade on outcome for canine rib chondrosarcomas.

A multi-feature spatial–temporal fusion network for traffic flow prediction

The traffic flow prediction is the key to alleviate traffic congestion, yet very challenging due to the complex influence factors. Currently, the most of deep learning models are designed to dig out the intricate dependency in continuous standardized sequences, which are dependent to high requirements for data continuity and regularized distribution. However, the data discontinuity and irregular distribution are inevitable in the real-world practical application, then we need find a way to utilize the powerful effect of the multi-feature fusion rather than continuous relation in standardized sequences. To this end, we conduct the prediction based on the multiple traffic features reflecting the complex influence factors. Firstly, we propose the ATFEM, an adaptive traffic features extraction mechanism, which can select important influence factors to construct joint temporal features matrix and global spatial features matrix according to the traffic condition. In this way, the feature’s representation ability can be improved. Secondly, we propose the MFSTN, a multi-feature spatial–temporal fusion network, which include the temporal transformer encoder and graph attention network to obtain the latent representation of spatial–temporal features. Especially, we design the scaled spatial–temporal fusion module, which can automatically learn optimal fusion weights, further adapt to inconsistent spatial–temporal dimensions. Finally, the multi-layer perceptron gets the mapping function between these comprehensive features and traffic flow. This method helps to improve the interpretability of the prediction. Experimental results show that the proposed model outperforms a variety of baselines, and it can accurately predict the traffic flow when the data missing rate is high.

Phonological Variation of /r/ in Selected Arabic Dialects

The present study offers a descriptive examination of the Arabic phoneme /r/, renowned for its intricate and diverse nature within the Arabic phonological system. This research investigates the varied manifestations of the phoneme /r/ in different regional dialects, with a specific emphasis on its pronunciation in the initial, middle, and final positions of words. The paper's significance is rooted in its contribution to the comprehension of the Arabic /r/ phoneme, a multifaceted and diverse sound within Arabic phonological systems. The paper illuminates the phonetic variability of the /r/ by analyzing its distinct manifestations across diverse Arabic dialects and regions. This can potentially facilitate language acquisition and instruction. The investigation conducted by the authors pertains to the marked and unmarked characteristics of the phonological /r/ in Arabic. The population under investigation comprises a randomly selected group of Arabic-speaking individuals aged 18 years and older, encompassing both genders and hailing from diverse countries and regions. The study employs a descriptive-analytical methodology, utilizing a unique features matrix and the Audacity Sound Recording and Analysis Application to examine the gathered data. The findings of the research enhance the comprehension of the phonetic variability that typifies the Arabic /r/ phoneme and elucidate its manifestation in diverse contexts and geographical locations. Furthermore, the outcomes of this investigation hold significance for the domain of language education and acquisition, and the techniques utilized were designed to guarantee the precision and dependability of the gathered information.

Improved XCT image automatic segmentation for quantitative characterization of the meso-morphological features in the damaged braided composite fabric

Accurately identifying various meso-morphological features and sub-phases within the damaged braided composite fabric is crucial for assessing the mechanical properties of composites under complex loading conditions and different processing parameters. However, obtaining 3D reconstructions that can be quantitatively analyzed is still a challenge, primarily due to the low contrast of the meso-morphological features (yarn tow and matrix crack networks). In this work, a new data enhancement algorithm is proposed to generate a realistic-looking artificial training dataset for enriching the information of the meso-morphological features. Then, the effect of various training networks and training parameters (the size of real and hybrid training dataset, number of epochs) on segmentation performance were examined. Finally, the meso-morphological features were statistically analyzed to accurately evaluate the mechanical properties of the composite material. The work presented here provides an effective tool that enables the quantitative analysis of the dynamic crack evolution process under the axial load of the composite materials.

RSO based Optimization of Random Forest Classifier for Fault Detection and Classification in Photovoltaic Arrays

Fault detection and classification in photovoltaic arrays are critical for increasing grid reliability and reducing the power losses. This paper assesses twelve machine learning classifiers for their effectiveness in detecting and classifying faults in Photovoltaic (PV) systems. Multiple validation methods were used for the algorithm evaluation, including K-fold, stratified K-fold, leave-one-out, and random split cross-validation approaches to ensure robust performance measures. The applied selection criterion of the top performing classifier are the accuracy, precision, recall, and computing efficiency. The utilized dataset, comprising samples with various fault kinds under diverse environmental conditions, received thorough preprocessing to enhance model training and assure generalizability. A large dataset of roughly 10,000 samples was utilized in this research for the model training and to run multiple random tests on new and unseen data. This dataset provides a fair representation of multiple fault types such as the healthy, Line to Line (LL), Line to Ground (LG), Partial Shading (PS), and Complete Shading faults (CS). The data preprocessing comprised normalization, handling of missing values by taking the average, and applying multiple statical analysis approaches to reduce the size of the features matrix and to improve the dependability of the model's predictions across varying operational circumstances. The results illustrate the best performance utilizing the optimized version of the Random Forest classifier, reaching an average fault detection accuracy of 100% and fault classification accuracy of 94.7%, the hyperparameters of the classifier was optimized using Random Search Optimization algorithm(RSO).

Collaborative Multi-Metadata Fusion to Improve the Classification of Lumbar Disc Herniation.

Computed tomography (CT) images are the most commonly used radiographic imaging modality for detecting and diagnosing lumbar diseases. Despite many outstanding advances, computer-aided diagnosis (CAD) of lumbar disc disease remains challenging due to the complexity of pathological abnormalities and poor discrimination between different lesions. Therefore, we propose a Collaborative Multi-Metadata Fusion classification network (CMMF-Net) to address these challenges. The network consists of a feature selection model and a classification model. We propose a novel Multi-scale Feature Fusion (MFF) module that can improve the edge learning ability of the network region of interest (ROI) by fusing features of different scales and dimensions. We also propose a new loss function to improve the convergence of the network to the internal and external edges of the intervertebral disc. Subsequently, we use the ROI bounding box from the feature selection model to crop the original image and calculate the distance features matrix. We then concatenate the cropped CT images, multiscale fusion features, and distance feature matrices and input them into the classification network. Next, the model outputs the classification results and the class activation map (CAM). Finally, the CAM of the original image size is returned to the feature selection network during the upsampling process to achieve collaborative model training. Extensive experiments demonstrate the effectiveness of our method. The model achieved 91.32% accuracy in the lumbar spine disease classification task. In the labelled lumbar disc segmentation task, the Dice coefficient reaches 94.39%. The classification accuracy in the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) reaches 91.82%.

The risk co-de model: detecting psychosocial processes of risk perception in natural language through machine learning

This paper presents a classification system (risk Co-De model) based on a theoretical model that combines psychosocial processes of risk perception, including denial, moral disengagement, and psychological distance, with the aim of classifying social media posts automatically, using machine learning algorithms. The risk Co-De model proposes four macro-categories that include nine micro-categories defining the stance towards risk, ranging from Consciousness to Denial (Co-De). To assess its effectiveness, a total of 2381 Italian tweets related to risk events (such as the Covid-19 pandemic and climate change) were manually annotated by four experts according to the risk Co-De model, creating a training set. Each category was then explored to assess its peculiarity by detecting co-occurrences and observing prototypical tweets classified as a whole. Finally, machine learning algorithms for classification (Support Vector Machine and Random Forest) were trained starting from a text chunks x (multilevel) features matrix. The Support Vector Machine model trained on the four macro-categories achieved an overall accuracy of 86% and a macro-average F1 score of 0.85, indicating good performance. The application of the risk Co-De model addresses the challenge of automatically identifying psychosocial processes in natural language, contributing to the understanding of the human approach to risk and informing tailored communication strategies.

Validating the credibility of solar simulation tools using a real-world case study

Solar energy is being recognized around the world for generating clean power using photovoltaic (PV) panels. Inadequate planning and sizing of PV panel systems leads to increased costs, emphasising the significance of reviewing solar energy simulationtools for accuracy and credibility. Even with the current literature, it could be challenging to select the software tools that are appropriate for the research. This paper identified, reviewed and validated four system-based (System Advisor Model (SAM), PVsyst, PV*Sol and Solarius PV) and three online software tools (Helioscope, PV-Watts and PV-GIS). It provides necessary information on the software tools which compiled in a features matrix under seven categories. The validation process involves using real-world data from an 84.5KW solar plant installed on a building in Nagpur, India, to assess the accuracy and performance of the software tools. The statistical metrics used for validation are Absolute Deviation (AD), Root Mean Square Error (RMSE) accompanied by Mean Bias Error (MBE) and Nash-Sutcliffe Efficiency (NSE) for a comprehensive understanding of the accuracy and reliability. The findings show a discrepancy in the simulated output data and the recorded data of the tools with PV*Sol and Solarius PV showcasing the least error. In conclusion, the findings of this research provide valuable insights and guidance for users seeking to make informed decisions in selecting the most appropriate Solar PV simulation tool tailored to their specific needs and requirements.

Natural gas pipeline leak detection based on acoustic signal analysis and feature reconstruction

The natural gas pipeline leakage detection task based on acoustic signal has some problems such as background noise coverage, lack of effective features, and low fault identification accuracy caused by small sample data. However, only one of these problems was usually studied in previous technologies. Almost no one has attempted to challenge multiple issues at the same time. In this study, a natural gas pipeline leak detection model that integrates acoustic feature processing techniques and feature reconstruction is proposed to resolve the above problems collaboratively. This model consists of two components. The first component is a feature processing technique of the acoustic signal that integrates frequency domain vector denoising and time domain associative function feature enhancement. The second component is a one-dimensional convolutional neural network with an expanded structural feature encoder (FAE) for feature reconstruction (FAE-1D-CNN). In the feature processing stage of the acoustic signal, firstly, the acoustic signal collected by the acoustic sensor is discretized into a digital signal. Secondly, the energy modal function is used to perform high/low energy modal clustering of digital signal features. The feature validity is enhanced by adding association factors to the low-energy modal features matrix. A low-pass filtering method is then used in the high-energy modal features to remove the background noise coverage of the high-frequency components. In the fault feature extraction stage, a feature encoder (FAE) is introduced in the 1D-CNN network to extract effective fault features while performing secondary reconstruction of local spatial features, addressing the problem of small sample leakage signals with few effective fault characteristics. The global average pooling layer is used instead of the fully connected layer, and the Softmax function is adopted as the classifier for fault discrimination. The performance of the proposed method was evaluated on the GPLA-12 dataset, and the fault identification accuracy is up to 95.17%. Compared with other competing methods, the method in this paper exhibits optimal performance and has broad application prospects.

Generalized Penalized Constrained Regression: Sharp Guarantees in High Dimensions with Noisy Features

The generalized penalized constrained regression (G-PCR) is a penalized model for high-dimensional linear inverse problems with structured features. This paper presents a sharp error performance analysis of the G-PCR in the over-parameterized high-dimensional setting. The analysis is carried out under the assumption of a noisy or erroneous Gaussian features matrix. To assess the performance of the G-PCR problem, the study employs multiple metrics such as prediction risk, cosine similarity, and the probabilities of misdetection and false alarm. These metrics offer valuable insights into the accuracy and reliability of the G-PCR model under different circumstances. Furthermore, the derived results are specialized and applied to well-known instances of G-PCR, including l1-norm penalized regression for sparse signal recovery and l2-norm (ridge) penalization. These specific instances are widely utilized in regression analysis for purposes such as feature selection and model regularization. To validate the obtained results, the paper provides numerical simulations conducted on both real-world and synthetic datasets. Using extensive simulations, we show the universality and robustness of the results of this work to the assumed Gaussian distribution of the features matrix. We empirically investigate the so-called double descent phenomenon and show how optimal selection of the hyper-parameters of the G-PCR can help mitigate this phenomenon. The derived expressions and insights from this study can be utilized to optimally select the hyper-parameters of the G-PCR. By leveraging these findings, one can make well-informed decisions regarding the configuration and fine-tuning of the G-PCR model, taking into consideration the specific problem at hand as well as the presence of noisy features in the high-dimensional setting.

Automatic Analysis of MRI Images for Early Prediction of Alzheimer's Disease Stages Based on Hybrid Features of CNN and Handcrafted Features.

Alzheimer's disease (AD) is considered one of the challenges facing health care in the modern century; until now, there has been no effective treatment to cure it, but there are drugs to slow its progression. Therefore, early detection of Alzheimer's is vital to take needful measures before it develops into brain damage which cannot be treated. Magnetic resonance imaging (MRI) techniques have contributed to the diagnosis and prediction of its progression. MRI images require highly experienced doctors and radiologists, and the analysis of MRI images takes time to analyze each slice. Thus, deep learning techniques play a vital role in analyzing a huge amount of MRI images with high accuracy to detect Alzheimer's and predict its progression. Because of the similarities in the characteristics of the early stages of Alzheimer's, this study aimed to extract the features in several methods and integrate the features extracted from more than one method into the same features matrix. This study contributed to the development of three methodologies, each with two systems, with all systems aimed at achieving satisfactory accuracy for the detection of AD and predicting the stages of its progression. The first methodology is by Feed Forward Neural Network (FFNN) with the features of GoogLeNet and DenseNet-121 models separately. The second methodology is by FFNN network with combined features between GoogLeNet and Dense-121 models before and after high-dimensionality reduction of features using the Principal Component Analysis (PCA) algorithm. The third methodology is by FFNN network with combined features between GoogLeNet and Dense-121 models separately and features extracted by Discrete Wavelet Transform (DWT), Local Binary Pattern (LBP) and Gray Level Co-occurrence Matrix (GLCM) methods called handcrafted features. All systems yielded super results in detecting AD and predicting the stages of its progression. With the combined features of the DenseNet-121 and handcrafted, the FFNN achieved an accuracy of 99.7%, sensitivity of 99.64%, AUC of 99.56%, precision of 99.63%, and a specificity of 99.67%.

Extractive Text Summarization Using Deep Learning for Tigrigna Language

With the ever-increasing amounts of textual material such as web pages, news articles, blogs, microblogs, and similar, the Internet became the massive body of unstructured information. In this paper to deal with the issues for the availability of more and more information with less time, the extractive text summarization using the deep learning model was used. In this paper, the proposed approach uses three basic stages of feature extraction, feature enhancement, and summary generation of the given news article to extract the core information, to produce well understandable summary and save reader’s time. In the feature extraction, We explore various features to improve the extracted sentences to the summary by the score and rank of the extracted features matrix by calculating the top thematic words, paragraph segmentation, sentences length & position, proper nouns, and TF-ISF, and the sum of the feature vector given to RBM to enhance the extracted feature vector and finally generate the final summarization by taking top high scores and 50% 0f the sum second higher scores from the enhanced feature extracted scores. For experimenting purpose, we have used 10 news articles from the total gathered news articles gathered from BBC-Tigrigna, Fana-Tigrigna and VOA-Tigrigna news website. The evaluation of the extracted summary was evaluated using Recall-Oriented Understudy for Gisting Evaluation (ROUGE) to compare the system extracted summary with the reference / manual summary prepared by human experts. According the experimentation, the average score of ROUG-1 shows 49% for recall, 39% precession, 42% for F-score and for the ROUGE-2 shows that 32% recall, 26% precession and 28% for F-score, for ROUGE-l also shows that 39% of recall, 33% of Precession, and 35% of F-scores. The result shows the proposed approach have higher result in Rouge-1 and the F-score or harmonic mean of precision and recall is 42% and it solves the problems of information overloading in the ever-increasing available news articles by generating the extractive summarizations.

Multisensor-Based Continuous Authentication of Smartphone Users With Two-Stage Feature Extraction

The one-time authentication mechanism in traditional authentication methods cannot continuously authenticate smartphone users’ identities throughout the session. Continuous authentication based on the behavioral biometrics recorded by the built-in sensors can solve this issue. However, the existing methods based on multisensor have poor ability to extract valuable features that can represent smartphone users’ behavioral patterns. This article proposes a novel method combining the manual construction and the deep metric learning method to perform two-stage feature extraction, respectively. We transform the time-series raw data from three sensors (accelerometer, gyroscope, and magnetometer) into 69 statistical features in the first stage. Furthermore, unlike the existing serial feature fusion methods, we innovatively fuse the constructed statistical features from three sensors into a three-channel matrix. Then, the fused features matrix with a three-channel is fed to the deep metric learning model for the second stage of feature extraction. We use the elliptic envelope algorithm to classify the user as a legitimate user or an impostor. Finally, we evaluate the performance of the proposed method on two public data sets. Experimental results show that our method can achieve an average accuracy of 99.71% and an average equal error rate (EER) of 0.56% on the hand movement, movement, orientation, and grasp data set, and an average accuracy of 99.59% and an average EER of 0.61% on the BrainRun data set.

Predicting long-term production dynamics in tight/shale gas reservoirs with dual-stage attention-based TEN-Seq2Seq model: A case study in Duvernay formation

Production dynamics forecasting plays an important role in the decision-making and development scenario evaluation process throughout the entire life cycle of the unconventional tight/shale gas reservoirs. The traditional method such as decline curve analysis can't be applied prior to the wells are put into production as it heavily depends on the historical production for the estimation of parameters. In this work, a new artificial intelligence framework is proposed to predict the well behaviors by simultaneously processing the sequential and tabular data including well depth, proppant tonnage, and fracturing stages. Specifically, a time evolution network is employed first to encode the tabular features matrix into a pseudo-sequence tensor, and then an encoder-decoder architecture based on the dual-stage attention mechanism is used to extract effective information from the encoded information and capture long-term dependencies relationship. A comparison of the proposed model with the fully connected neural network (FCNN) and the long and short-term memory (LSTM) network indicates that the new framework has better generalization performance and robustness to predict well productivities, that is, the prediction errors are reduced by 65% and 50% respectively compared with LSTM and FCNN. Moreover, a bidirectional parametric rectified linear unit (BPReLU) is employed to adaptively learn the sign and magnitude of slopes. It is found that the error is further reduced by approximately 10% compared to that using PReLU. Also, four different target variables are defined, and the experimental results reveal that the average rate within the production time Vi is much easier to predict, with an average error of 19%.

A novel cancelable fingerprint scheme based on random security sampling mechanism and relocation bloom filter

The existing cancellable fingerprint template based on the bloom filter either suffers from a lack of template security or low recognition performance. To solve the above problems, the paper proposes a novel cancelable fingerprint protection scheme based on a random sampling mechanism and relocation bloom filter. The security matrix is generated by a random sampling mechanism to increase the security and recognition accuracy of the scheme. The structure of a random sampling mechanism is as follows. Firstly, we use the key management system to generate keys and a chaotic matrix with randomness. Secondly, the key is used as a random initial address to extract overlapping feature sub-blocks from fingerprint features. Then the feature sub-blocks are combined to generate a feature matrix. Therefore, the feature matrix has richer fingerprint information than the original fingerprint features. Finally, a security matrix is generated by the XOR fusion of features matrix and chaotic matrix. Furthermore, we designed a relocation bloom filter with a parallel structure, which provides a way to avoid the decrease in performance caused by hash conflicts. What is more, it increased the security of the scheme that the relocatable bloom uses an irreversible function mapping function to transform the security matrix into a cancelable fingerprint template. The experimental results demonstrate that this scheme has state-of-the-art accuracy performance on benchmark FVC2002 and FVC2004 fingerprint databases. The security analysis shows that the scheme meets the cancellable biometrics protection scheme criteria and has higher security.

Variational Graph Autoencoder with Mutual Information Maximization for Graph Representations Learning

Graph neural network (GNN) is a powerful representation learning framework for graph-structured data. Some GNN-based graph embedding methods, including variational graph autoencoder (VGAE), have been presented recently. However, existing VGAE-based methods typically focus on reconstructing the adjacent matrix, i.e. topological structure, instead of the node features matrix, this strategy makes graphical features difficult to be fully learned, which weakens and restricts the capacity of a generative network to learn higher-quality representations. To address the issue, we use a contrastive estimator on the representation mechanism, i.e. on the encoding process under the framework of VGAE. In particular, we maximize the mutual information (MI) between encoded latent representation and node attributes which acts as a regularizer forcing the encoder to select the most informative with respect to the node attributes. Additionally, we also solve another key question how to effectively estimate the mutual information by drawing samples from the joint and marginal, and explain why the maximization of MI can contribute to the encoder obtaining more node feature information. Ultimately, extensive experiments on three citation networks and four web-age networks show that our method outperforms contemporary popular algorithms (such as DGI) on node classifications and clustering tasks, and the best result is an [Formula: see text] increase on node clustering task.

Characterization and chemometric modelling of mechanically recycled polypropylene for automotive manufacturing

In order to successfully apply recycled plastics for car manufacturing, it is necessary to identify chemical and physical changes during the recycling process to ensure quality and assess potential risks of material deterioration and failure. For a common automotive plastic material such as polypropylene (PP), expected consequences of repeated processing are polymer chain scission due to thermal and mechanical stress, oxidation, contamination, and changes in the composition of additives. In this study, we describe a systematic chemometric approach towards quantitative prediction models for a target value, exemplified by the recyclate content in PP. Based on a multimodal material analysis, a feature matrix is composed from large, heterogeneous datasets through dimensionality reduction. Analytical methods include infrared spectroscopy, Raman microscopy, combined thermodesorption-gas chromatography-mass spectrometry, as well as size exclusion and high performance liquid chromatography. Analytical features and prediction models are then selected and scored. We show that the weight content of recyclate can be predicted with an accuracy ∼2% using simple linear models. We present an outlook for model deployment to the factory floor by using fast, non-destructive analytics such as IR spectroscopy. • Comparison of commercially available virgin and recycled PP by chemical analysis. • Multivariate analysis of spectra using non-negative matrix factorization. • Degradation of stabilizers and decrease of average molecular weight in the recyclate. • Combining datasets from different analytical techniques in a features matrix. • Testing and scoring different feature combinations and regression models to predict a target value.

Enhanced Epileptic Seizure diagnosis using EEG Signals with Support vector machine and Bagging Classifiers

Many approaches have been proposed using Electroencephalogram (EEG) to detect epilepsy seizures in their early stages. Epilepsy seizure is a severe neurological disease. Practitioners continue to rely on manual testing of EEG signals. Artificial intelligence (AI) and Machine Learning (ML) can effectively deal with this problem. ML can be used to classify EEG signals employing feature extraction techniques. This work focuses on automated detection for epilepsy seizures using ML techniques. Various algorithms are investigated, such as Bagging, Decision Tree (DT), Adaboost, Support vector machine (SVM), K-nearest neighbors(KNN), Artificial neural network(ANN), Naïve Bayes, and Random Forest (RF) to distinguish injected signals from normal ones with high accuracy. In this work, 54 Discrete wavelet transforms (DWTs) are used for feature extraction, and the similarity distance is applied to identify the most powerful features. The features are then selected to form the features matrix. The matrix is subsequently used to train ML. The proposed approach is evaluated through different metrics such as F-measure, precision, accuracy, and Recall. The experimental results show that the SVM and Bagging classifiers in some data set combinations, outperforming all other classifiers

Noninvasive Method for Predicting the Expression of Ki67 and Prognosis in Non-Small-Cell Lung Cancer Patients: Radiomics.

Purpose In this study, we aimed to develop and validate a noninvasive method based on radiomics to evaluate the expression of Ki67 and prognosis of patients with non-small-cell lung cancer (NSCLC). Patients and Methods. A total of 120 patients with NSCLC were enrolled in this retrospective study. All patients were randomly assigned to a training dataset (n = 85) and test dataset (n = 35). According to the preprocessed F-FDG PET/CT image of each patient, a total of 384 radiomics features were extracted from the segmentation of regions of interest (ROIs). The Spearman correlation test and least absolute shrinkage and selection operator (LASSO), after normalization on the features matrix, were applied to reduce the dimensionality of the features. Furthermore, multivariable logistic regression analysis was used to propose a model for predicting Ki67. The survival curve was used to explore the prognostic significance of radiomics features. Results A total of 62 Ki67 positive patients and 58 Ki67 negative patients formed the training set and test training dataset and test dataset. Radiomics signatures showed good performance in predicting the expression of Ki67 with AUCs of 0.86 (training dataset) and 0.85 (test dataset). Validation and calibration showed that the radiomics had a strong predictive power in patients with NSCLC survival, which was significantly close to the effect of Ki67 expression on the survival of patients with NSCLC. Conclusion Radiomics signatures based on preoperative F-FDG PET/CT could distinguish the expression of Ki67, which also had a strong predictive performance for the survival outcome.

Time series classification based on complex network

Time series classification is an important topic in data mining. Time series classification methods have been studied by many researchers. A feature-weighted classification method is proposed based on complex network. There are four stages in the proposed classification algorithm. First, we obtain visible points and edge weights by applying an improved weighted visibility graph algorithm to transform the original time series into weighted complex networks. Next, weighted features are calculated based on visible points and weights, and unable-weighted features are obtained by utilizing NetworkX. After that, we combine two types of features to obtain a total features matrix. We then reconsider the weight of each independent feature by utilizing the importance function of random forest to measure the relative importance of features and construct a importance weighted features matrix. Finally, we utilize a traditional classifier called random forest to cluster the weighted features matrix and generate classification results. Simultaneously, a novel feature weight calculation method is applied during the classification process. We compare the proposed method to other classification methods and the results indicate that the proposed method can improve classification accuracy for time series datasets.