Vocal Cord Leukoplakia Classification Using Siamese Network Under Small Samples of White Light Endoscopy Images

OBJECTIVE

Accurate vocal cord leukoplakia classification is instructive for clinical diagnosis and surgical treatment. This article introduces a reliable very deep Siamese network for accurate vocal cord leukoplakia classification.

STUDY DESIGN

A study of a classification network based on a retrospective database.

SETTING

Academic university and hospital.

METHODS

The white light image datasets of vocal cord leukoplakia used in this article were classified into 6 classes: normal tissues, inflammatory keratosis, mild dysplasia, moderate dysplasia, severe dysplasia, and squamous cell carcinoma. The classification performance was assessed by comparing it with 6 classical deep learning models, including AlexNet, VGG Net, Google Inception, ResNet, DenseNet, and Vision Transformer.

RESULTS

Experiments show the superior classification performance of our proposed network compared to state-of-the-art methods. The overall accuracy is 0.9756. The values of sensitivity and specificity are very high as well. The confusion matrix provides information for the 6-class classification task and demonstrates the superiority of our proposed network.

CONCLUSION

Our very deep Siamese network can provide accurate classification results of vocal cord leukoplakia, which facilitates early detection, clinical diagnosis, and surgical treatment. The excellent performance obtained in white light images can reduce the cost for patients, especially those living in developing countries.

SST: a snore shifted-window transformer method for potential obstructive sleep apnea patient diagnosis

Objective.Obstructive sleep apnea (OSA) is a high-incidence disease that is seriously harmful and potentially dangerous. The objective of this study was to develop a noncontact sleep audio signal-based method for diagnosing potential OSA patients, aiming to provide a more convenient diagnostic approach compared to the traditional polysomnography (PSG) testing.Approach.The study employed a shifted window transformer model to detect snoring audio signals from whole-night sleep audio. First, a snoring detection model was trained on large-scale audio datasets. Subsequently, the deep feature statistical metrics of the detected snore audio were used to train a random forest classifier for OSA patient diagnosis.Main results.Using a self-collected dataset of 305 potential OSA patients, the proposed snore shifted-window transformer method (SST) achieved an accuracy of 85.9%, a sensitivity of 85.3%, and a precision of 85.6% in OSA patient classification. These values surpassed the state-of-the-art method by 9.7%, 10.7%, and 7.9%, respectively.Significance.The experimental results demonstrated that SST significantly improved the noncontact audio-based OSA diagnosis performance. The study's findings suggest a promising self-diagnosis method for potential OSA patients, potentially reducing the need for invasive and inconvenient diagnostic procedures.

Vocal cord leukoplakia classification using deep learning models in white light and narrow band imaging endoscopy images

BACKGROUND

Accurate vocal cord leukoplakia classification is critical for the individualized treatment and early detection of laryngeal cancer. Numerous deep learning techniques have been proposed, but it is unclear how to select one to apply in the laryngeal tasks. This article introduces and reliably evaluates existing deep learning models for vocal cord leukoplakia classification.

METHODS

We created white light and narrow band imaging (NBI) image datasets of vocal cord leukoplakia which were classified into six classes: normal tissues (NT), inflammatory keratosis (IK), mild dysplasia (MiD), moderate dysplasia (MoD), severe dysplasia (SD), and squamous cell carcinoma (SCC). Vocal cord leukoplakia classification was performed using six classical deep learning models, AlexNet, VGG, Google Inception, ResNet, DenseNet, and Vision Transformer.

RESULTS

GoogLeNet (i.e., Google Inception V1), DenseNet-121, and ResNet-152 perform excellent classification. The highest overall accuracy of white light image classification is 0.9583, while the highest overall accuracy of NBI image classification is 0.9478. These three neural networks all provide very high sensitivity, specificity, and precision values.

CONCLUSION

GoogLeNet, ResNet, and DenseNet can provide accurate pathological classification of vocal cord leukoplakia. It facilitates early diagnosis, providing judgment on conservative treatment or surgical treatment of different degrees, and reducing the burden on endoscopists.

Application and interpretation of machine learning models in predicting the risk of severe obstructive sleep apnea in adults

BACKGROUND

Obstructive sleep apnea (OSA) is a globally prevalent disease with a complex diagnostic method. Severe OSA is associated with multi-system dysfunction. We aimed to develop an interpretable machine learning (ML) model for predicting the risk of severe OSA and analyzing the risk factors based on clinical characteristics and questionnaires.

METHODS

This was a retrospective study comprising 1656 subjects who presented and underwent polysomnography (PSG) between 2018 and 2021. A total of 23 variables were included, and after univariate analysis, 15 variables were selected for further preprocessing. Six types of classification models were used to evaluate the ability to predict severe OSA, namely logistic regression (LR), gradient boosting machine (GBM), extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), bootstrapped aggregating (Bagging), and multilayer perceptron (MLP). All models used the area under the receiver operating characteristic curve (AUC) was calculated as the performance metric. We also drew SHapley Additive exPlanations (SHAP) plots to interpret predictive results and to analyze the relative importance of risk factors. An online calculator was developed to estimate the risk of severe OSA in individuals.

RESULTS

Among the enrolled subjects, 61.47% (1018/1656) were diagnosed with severe OSA. Multivariate LR analysis showed that 10 of 23 variables were independent risk factors for severe OSA. The GBM model showed the best performance (AUC = 0.857, accuracy = 0.766, sensitivity = 0.798, specificity = 0.734). An online calculator was developed to estimate the risk of severe OSA based on the GBM model. Finally, waist circumference, neck circumference, the Epworth Sleepiness Scale, age, and the Berlin questionnaire were revealed by the SHAP plot as the top five critical variables contributing to the diagnosis of severe OSA. Additionally, two typical cases were analyzed to interpret the contribution of each variable to the outcome prediction in a single patient.

CONCLUSIONS

We established six risk prediction models for severe OSA using ML algorithms. Among them, the GBM model performed best. The model facilitates individualized assessment and further clinical strategies for patients with suspected severe OSA. This will help to identify patients with severe OSA as early as possible and ensure their timely treatment.

TRIAL REGISTRATION

Retrospectively registered.

A shallow mirror transformer for subject-independent motor imagery BCI

OBJECTIVE

Motor imagery BCI plays an increasingly important role in motor disorders rehabilitation. However, the position and duration of the discriminative segment in an EEG trial vary from subject to subject and even trial to trial, and this leads to poor performance of subject-independent motor imagery classification. Thus, determining how to detect and utilize the discriminative signal segments is crucial for improving the performance of subject-independent motor imagery BCI.

APPROACH

In this paper, a shallow mirror transformer is proposed for subject-independent motor imagery EEG classification. Specifically, a multihead self-attention layer with a global receptive field is employed to detect and utilize the discriminative segment from the entire input EEG trial. Furthermore, the mirror EEG signal and the mirror network structure are constructed to improve the classification precision based on ensemble learning. Finally, the subject-independent setup was used to evaluate the shallow mirror transformer on motor imagery EEG signals from subjects existing in the training set and new subjects.

MAIN RESULTS

The experiments results on BCI Competition IV datasets 2a and 2b and the OpenBMI dataset demonstrated the promising effectiveness of the proposed shallow mirror transformer. The shallow mirror transformer obtained average accuracies of 74.48% and 76.1% for new subjects and existing subjects, respectively, which were highest among the compared state-of-the-art methods. In addition, visualization of the attention score showed the ability of discriminative EEG segment detection. This paper demonstrated that multihead self-attention is effective in capturing global EEG signal information in motor imagery classification.

SIGNIFICANCE

This study provides an effective model based on a multihead self-attention layer for subject-independent motor imagery-based BCIs. To the best of our knowledge, this is the shallowest transformer model available, in which a small number of parameters promotes the performance in motor imagery EEG classification for such a small sample problem.

Overlapping filter bank convolutional neural network for multisubject multicategory motor imagery brain-computer interface

BACKGROUND

Motor imagery brain-computer interfaces (BCIs) is a classic and potential BCI technology achieving brain computer integration. In motor imagery BCI, the operational frequency band of the EEG greatly affects the performance of motor imagery EEG recognition model. However, as most algorithms used a broad frequency band, the discrimination from multiple sub-bands were not fully utilized. Thus, using convolutional neural network (CNNs) to extract discriminative features from EEG signals of different frequency components is a promising method in multisubject EEG recognition.

METHODS

This paper presents a novel overlapping filter bank CNN to incorporate discriminative information from multiple frequency components in multisubject motor imagery recognition. Specifically, two overlapping filter banks with fixed low-cut frequency or sliding low-cut frequency are employed to obtain multiple frequency component representations of EEG signals. Then, multiple CNN models are trained separately. Finally, the output probabilities of multiple CNN models are integrated to determine the predicted EEG label.

RESULTS

Experiments were conducted based on four popular CNN backbone models and three public datasets. And the results showed that the overlapping filter bank CNN was efficient and universal in improving multisubject motor imagery BCI performance. Specifically, compared with the original backbone model, the proposed method can improve the average accuracy by 3.69 percentage points, F1 score by 0.04, and AUC by 0.03. In addition, the proposed method performed best among the comparison with the state-of-the-art methods.

CONCLUSION

The proposed overlapping filter bank CNN framework with fixed low-cut frequency is an efficient and universal method to improve the performance of multisubject motor imagery BCI.

A Fault-Diagnosis Method for Railway Turnout Systems Based on Improved Autoencoder and Data Augmentation

In recent years, with the rapid increase in coverage and lines, security maintenance has become one of the top concerns with regard to railway transportation in China. As the key transportation infrastructure, the railway turnout system (RTS) plays a vital role in transportation, which will cause incalculable losses when accidents occur. The traditional fault-diagnosis and maintenance methods of the RTS are no longer applicable to the growing amount of data, so intelligent fault diagnosis has become a research hotspot. However, the key challenge of RTS intelligent fault diagnosis is to effectively extract the deep features in the signal and accurately identify failure modes in the face of unbalanced datasets. To solve the above two problems, this paper focuses on unbalanced data and proposes a fault-diagnosis method based on an improved autoencoder and data augmentation, which realizes deep feature extraction and fault identification of unbalanced data. An improved autoencoder is proposed to smooth the noise and extract the deep features to overcome the noise fluctuation caused by the physical characteristics of the data. Then, synthetic minority oversampling technology (SMOTE) is utilized to effectively expand the fault types and solve the problem of unbalanced datasets. Furthermore, the health state is identified by the Softmax regression model that is trained with the balanced characteristics data, which improves the diagnosis precision and generalization ability. Finally, different experiments are conducted on a real dataset based on a railway station in China, and the average diagnostic accuracy reaches 99.13% superior to other methods, which indicates the effectiveness and feasibility of the proposed method.

Railway accident prediction strategy based on ensemble learning

Railway accident prediction is of great significance for establishing an early warning mechanism and preventing the occurrences of accidents. Safety agencies rely on prediction models to design railroad risk management strategies. Based on historical railway accident data, an ensemble learning strategy for accident prediction is proposed. Firstly, an improved K-nearest neighbors (KNN) data imputation algorithm is proposed to solve the problem of missing data in the dataset. Then, to reduce the impact of imbalanced data on prediction performance, an AdaBoost-Bagging method is presented. Finally, according to the feature importance in the prediction model, accident features are ranked to identify new insights into the cause of the accident. The AdaBoost-Bagging prediction method is applied to the Federal Railroad Administration (FRA) dataset. The application results show that, compared with Artificial Neural Network (ANN), XGBoost, GBDT, Stacking and AdaBoost methods, AdaBoost-Bagging method has a smaller prediction error and faster inference time in predicting railway accidents. Accuracy, Precision, Recall and F1-score are 0.879, 0.879, 0.883 and 0.881 respectively, and the inference time is reduced by 23.38%, 12.15%, 6.66%, 3.17% and 11.41% respectively. The prediction method can well mine important features of railway accidents without knowing the accident mechanism or the relationship between various railway accidents and factors, e.g., the critic risk factors related to derailment and collision accidents are investigated in the prediction. The findings will be helpful to the prevention and management of railway accidents.

Transductive Semisupervised Deep Hashing

Deep hashing methods have shown their superiority to traditional ones. However, they usually require a large amount of labeled training data for achieving high retrieval accuracies. We propose a novel transductive semisupervised deep hashing (TSSDH) method which is effective to train deep convolutional neural network (DCNN) models with both labeled and unlabeled training samples. TSSDH method consists of the following four main ingredients. First, we extend the traditional transductive learning (TL) principle to make it applicable to DCNN-based deep hashing. Second, we introduce confidence levels for unlabeled samples to reduce adverse effects from uncertain samples. Third, we employ a Gaussian likelihood loss for hash code learning to sufficiently penalize large Hamming distances for similar sample pairs. Fourth, we design the large-margin feature (LMF) regularization to make the learned features satisfy that the distances of similar sample pairs are minimized and the distances of dissimilar sample pairs are larger than a predefined margin. Comprehensive experiments show that the TSSDH method can produce superior image retrieval accuracies compared to the representative semisupervised deep hashing methods under the same number of labeled training samples.

Algorithm Contest of Calibration-free Motor Imagery BCI in the BCI Controlled Robot Contest in World Robot Contest 2021: A survey

Objective:

From September 10 to 13, 2021, the finals of the BCI Controlled Robot Contest in World Robot Contest 2021 were held in Beijing, China. Eleven teams participated in the Algorithm Contest of Calibration-free Motor Imagery BCI. The participants employed both traditional electroencephalograph (EEG) analysis methods and deep learning-based methods in the contest. In this paper, we reviewed the algorithms utilized by the participants, extracted the trends and highlighted interesting approaches from these methods to inform future contests and research recommendations.

Method:

First, we analyzed the algorithms in separate steps, including EEG channel and signal segment setup, prepossessing technology, and classification model. Then, we emphasized the highlights of each algorithm. Finally, we compared the competition algorithm with the SOTA algorithm.

Results:

The algorithm employed in the finals performed better than that of the SOTA algorithm. During the final stage of the contest, four of the top five teams used convolutional neural network models, suggesting that with the rapid development of deep learning, convolutional neural network-based models have been the most popular methods in the field of motor imagery BCI.

Trusted Data Storage Architecture for National Infrastructure

National infrastructure is a material engineering facility that provides public services for social production and residents' lives, and a large-scale complex device or system is used to ensure normal social and economic activities. Due to the problems of difficult data collection, long project period, complex data, poor security, difficult traceability and data intercommunication, the archives management of most national infrastructure is still in the pre-information era. To solve these problems, this paper proposes a trusted data storage architecture for national infrastructure based on blockchain. This consists of real-time collection of national infrastructure construction data through sensors and other Internet of Things devices, conversion of heterogeneous data source data into a unified format according to specific business flows, and timely storage of data in the blockchain to ensure data security and persistence. Knowledge extraction of data stored in the chain and the data of multiple regions or fields are jointly modeled through federal learning. The parameters and results are stored in the chain, and the information of each node is shared to solve the problem of data intercommunication.

An efficient and global interactive optimization methodology for path planning with multiple routing constraints

Path planning problem is attracting wide attention in autonomous system and process industry system. The existed research mainly focuses on finding the shortest path from the source vertex to the termination vertex under loose constraints of vertex and edge. However, in realistic, the constraints such as specified vertexes, specified paths, forbidden paths and forbidden vertexes have to be considered, which makes the existing algorithms inefficient even infeasible. Aiming at solving the problems of complex path planning with multiple routing constraints, this paper organizes transforms the constraints into appropriate mathematical analytic expressions. Then, in order to overcome the defects of existing coding and optimization algorithms, an adaptive strategy for the vertex priority is proposed in coding, and an efficient and global optimization methodology based on swarm intelligence algorithms is put forward, which can make full use of the high efficiency of the local optimization algorithm and the high search ability of the global optimization algorithm. Moreover, the optimal convergence condition of the methodology is proved theoretically. Finally, two experiments are inducted, and the results demonstrated its efficiency and superiority.

Multi-model train state estimation based on multi-sensor parallel fusion filtering

Accurately determining a train's state is essential for passenger safety, operation efficiency, and maintenance. However, the actual operation state of a train is composed of a variety of modes and is disturbed by several known or unknown factors, for which an accurate estimator is required. Hence, in this paper, a train multi-mode model considering the actual operation environment is established, and a train state estimation method based on multi-sensor parallel fusion filter is proposed. In the parallel fusion filter, the current mode of train is determined by the proposed sliding window error and voting mechanism, and the global filter are constituted by the local filters, which are fused by linear-weighted summation. The simulation results demonstrate the effectiveness of our method in estimating the train's state. It is worth noting that even if monitoring data are missing or are abnormal, the state estimation accuracy of the proposed technique still meets the requirements of a real system, and the effectiveness and robustness of the method can be verified.

Parallel anomaly detection algorithm for cybersecurity on the highspeed train control system

With the rapid development of the high-speed train industry, the high-speed train control system has now been exposed to a complicated network environment full of dangers. This paper provides a speculative parallel data detection algorithm to rapidly detect the potential threats and ensure data transmission security in the railway network. At first, the structure of the high-speed train control data received by the railway control center was analyzed and divided tentatively into small chunks to eliminate the inside dependencies. Then the traditional threat detection algorithm based on deterministic finite automaton was reformed by the speculative parallel optimization so that the inline relationship's influences that affected the data detection order could be avoided. At last, the speculative parallel detection algorithm would inspect the divided data chunks on a distributed platform. With the help of both the speculative parallel technique and the distributed platform, the detection deficiency for train control data was improved significantly. The results showed that the proposed algorithm exhibited better performance and scalability when compared with the traditional, non-parallel detection method, and massive train control data could be inspected and processed promptly. Now it has been proved by practical use that the proposed algorithm was stable and reliable. Our local train control center was able to quickly detect the anomaly and make a fast response during the train control data transmission by adopting the proposed algorithm.

Prediction of Cancer-Related piRNAs Based on Network-Based Stratification Analysis

PIWI-interacting RNA (PiRNA) was discovered in 2006 and is expected to become a new biomarker for diagnosis and prognosis of various diseases. The purpose of this study is to explore functions of piRNAs and identify cancer subtypes on the basis of the pattern of transcriptome and somatic mutation data. A total of 285 510 SNPs in piRNAs and genes, which might affect piRNA biogenesis or piRNA targets binding were identified. Significant co-expression networks of piRNAs were then constructed separately for 12 major types of cancer. Finally, mutational matrices were mapped to piRNA network, propagated, and clustered for identification of cancer-related piRNAs and cancer subtypes. Findings showed that subtypes of three types of cancer (COAD, STAD and UCEC), which are significantly associated with survival were identified. Analysis of differentially expressed piRNAs in UCEC subtypes showed that piRNA function is closely related to cancer hallmarks “Enabling Replicative Immortality” and contributes to initiation of cancer.

An intelligent fault diagnosis method based on curve segmentation and SVM for rail transit turnout

With the development of intelligent transportation system, the maintenance of railway turnout is an essential daily task which was required to be efficiency and automatically. This paper presents an intelligent diagnosis method based on deep learning curve segmentation and the Support Vector Machine. Firstly, we studied the curve segmentation approach of the real-time monitoring power data collected form turnout, for which is an essential step and do a great help to improve the diagnose accuracy. Then based on the well pre-processed data sets, the SVM algorithm was applied to classify the samples and report the health states of the turnout which under testing. At last, the experiments were taken on the power data curve collected from the real turnouts, during which we compared the new diagnose method with conventional ones, and the results showed that the diagnose accuracy of proposed method can averaged to 98.5%. Compared with traditional SVM based frameworks, the proposed diagnosis method dramatically improves the accuracy which is more suitable for railway turnout.

Improving the performance of multisubject motor imagery-based BCIs using twin cascaded softmax CNNs

OBJECTIVE

Motor imagery (MI) EEG signals vary greatly among subjects, so scholarly research on motor imagery-based brain-computer interfaces (BCIs) has mainly focused on single-subject systems or subject-dependent systems. However, the single-subject model is applicable only to the target subject, and the small sample number greatly limits the performance of the model. This paper aims to study a convolutional neural network to achieve an adaptable MI-BCI that is applicable to multiple subjects.

APPROACH

In this paper, a twin cascaded softmax convolutional neural network (TCSCNN) is proposed for multisubject MI-BCIs. The proposed TCSCNN is independent and can be applied to any single-subject MI classification CNN model. First, to reduce the influence of individual differences, subject recognition and MI recognition are accomplished simultaneously. A cascaded softmax structure consisting of two softmax layers, related to subject recognition and MI recognition, is subsequently applied. Second, to improve the MI classification precision, a twin network structure is proposed on the basis of ensemble learning. TCSCNN is built by combining a cascaded softmax structure and twin network structure.

MAIN RESULTS

Experiments were conducted on three popular CNN models (EEGNet and Shallow ConvNet and Deep ConvNet from EEGDecoding) and three public datasets (BCI Competition IV datasets 2a and 2b and the High-Gamma dataset) to verify the performance of the proposed TCSCNN. The results show that compared with the state-of-the-art CNN model, the proposed TCSCNN obviously improves the precision and convergence of multisubject MI recognition.

SIGNIFICANCE

This study provides a promising scheme for multisubject MI-BCI, reflecting the progress made in the development and application of MI-BCIs.

A novel deep feature transfer-based OSA detection method using sleep sound signals

OBJECTIVE

Polysomnography is typically used to evaluate the severity of obstructive sleep apnea (OSA) but the inconvenience of application and high cost considerably affect the diagnostics. In this study, sleep sound signals are used to detect OSA in patients.

APPROACH

A deep feature transfer-based OSA detection approach is proposed. First, a deep convolutional neural network is trained on large-scale labeled audio data sets to distinguish respiration sounds from environmental noise. Second, the trained model is transferred to recognize respiration sounds in sleep sound signals. Third, the deep features of the detected respiration sounds are used to train a logistic regression classifier to identify OSA patients from potential patients. Polysomnography-based diagnosis is used as a reference.

MAIN RESULTS

A self-collected data set of 132 potential OSA patients is applied in OSA detection experiments. The OSA detection performances are tested on four models for different apnea-hypopnea index thresholds and sexes resulting in accuracies of 80.17%, 80.21%, 81.63% and 77.22%. The corresponding areas under the receiver operating characteristic curves are 0.82, 0.80, 0.81 and 0.79. In addition, the proposed method presented a significant performance improvement compared with the state-of-the-art methods.

SIGNIFICANCE

Big data, deep learning and transfer learning can be successfully applied to improve diagnostic accuracy in OSA detection. The performance of the proposed approach is superior to that of traditional audio analysis technology. The proposed method significantly reduces difficulties in OSA detection and diagnosis, such that potential OSA patients can perform initial inspections by themselves at home.

A Memetic Particle Swarm Optimization Algorithm for Community Detection in Complex Networks

In recent years, community detection has become a hot research topic in complex networks. Many of the proposed algorithms are for detecting community based on the modularity Q. However, there is a resolution limit problem in modularity optimization methods. In order to detect the community structure more effectively, a memetic particle swarm optimization algorithm (MPSOA) is proposed to optimize the modularity density by introducing particle swarm optimization-based global search operator and tabu local search operator, which is useful to keep a balance between diversity and convergence. For comparison purposes, two state-of-the-art algorithms, namely, meme-net and fast modularity, are carried on the synthetic networks and other four real-world network problems. The obtained experiment results show that the proposed MPSOA is an efficient heuristic approach for the community detection problems.

Asymmetric double-image encryption based on cascaded discrete fractional random transform and logistic maps

A double-image encryption is proposed based on the discrete fractional random transform and logistic maps. First, an enlarged image is composited from two original images and scrambled in the confusion process which consists of a number of rounds. In each round, the pixel positions of the enlarged image are relocated by using cat maps which are generated based on two logistic maps. Then the scrambled enlarged image is decomposed into two components. Second, one of two components is directly separated into two phase masks and the other component is used to derive the ciphertext image with stationary white noise distribution by using the cascaded discrete fractional random transforms generated based on the logistic map. The cryptosystem is asymmetric and has high resistance against to the potential attacks such as chosen plaintext attack, in which the initial values of logistic maps and the fractional orders are considered as the encryption keys while two decryption keys are produced in the encryption process and directly related to the original images. Simulation results and security analysis verify the feasibility and effectiveness of the proposed encryption scheme.