NETWORK ANOMALIES DETECTION APPROACH BASED ON WEIGHTED VOTING

To avoid information systems malfunction, their integrity disruption, availability violation as well as data confidentiality, it is necessary to detect anomalies in information system operation as quickly as possible. The anomalies are usually caused by malicious activity – information systems attacks. However, the current approaches to detect anomalies in information systems functioning have never been perfect. In particular, statistical and signature-based techniques do not allow detection of anomalies based on modifications of well-known attacks, dynamic approaches based on machine learning techniques result in false responses and frequent anomaly miss-outs. Therefore, various hybrid solutions are being frequently offered on the basis of those two approaches. The paper suggests a hybrid approach to detect anomalies by combining computationally efficient classifiers of machine learning with accuracy increase due to weighted voting. Pilot evaluation of the developed approach proved its feasibility for anomaly detection systems.

Improved Selection of Rare Reactions in Template-Based Retrosynthesis Predictions

Identifying synthetic routes for molecules of interest is a crucial step when discovering new drugs or materials. To find synthetic routes, we can use computer-assisted synthesis planning using expansion policy networks trained on reaction templates extracted from patents and the literature. However, experience has shown that these networks are biased towards frequently reported reactions. This study shows that changing the molecular representation from an extended-connectivity fingerprint to a simple graph representation can increase the accuracy for templates used less than five times by 5.0- 8.5% points. We also illustrate that a simple oversampling of the training set yielded a top-1 accuracy increase in the 17-20% point range for templates used five times or less.

Methods and means for real-time object recognition accuracy increase in video images on ios mobile platform

As a result of the analytical review, it was established that the family of Yolo models is a promising area of search and recognition of objects. However, existing implementations do not support the ability to run the model on the iOS platform. To achieve these goals, a comprehensive scalable conversion system has been developed to improve the recognition accuracy of arbitrary models based on the Docker system. The method of improvement is to add a layer with the Mish activation function to the original model. The method of conversion is to quickly convert any Yolo model to CoreML format. As part of the study of these techniques, a model of the neural network Yolov4_TCAR was created. Additionally, a method of accelerating the load on the CPU using an additional layer of neural network with the function of activating Mish in Swift for the iOS mobile platform was added. As a result, the effectiveness of the Mish activation function, the CPU load of the mobile device, the amount of RAM used, and the frame rate when using the improved original Yolov4-TCAR model were studied. The results of the research confirmed the functioning of the algorithm for conversion and accuracy increase of the neural network model in real-time.

Graph Analysis of EEG Functional Connectivity Networks During a Letter-Speech Sound Binding Task in Adult Dyslexics

We performed an EEG graph analysis on data from 31 typical readers (22.27 ± 2.53 y/o) and 24 dyslexics (22.99 ± 2.29 y/o), recorded while they were engaged in an audiovisual task and during resting-state. The task simulates reading acquisition as participants learned new letter-sound mappings via feedback. EEG data was filtered for the delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz) bands. We computed the Phase Lag Index (PLI) to provide an estimate of the functional connectivity between all pairs of electrodes per band. Then, networks were constructed using a Minimum Spanning Tree (MST), a unique sub-graph connecting all nodes (electrodes) without loops, aimed at minimizing bias in between groups and conditions comparisons. Both groups showed a comparable accuracy increase during task blocks, indicating that they correctly learned the new associations. The EEG results revealed lower task-specific theta connectivity, and lower theta degree correlation over both rest and task recordings, indicating less network integration in dyslexics compared to typical readers. This pattern suggests a role of theta oscillations in dyslexia and may reflect differences in task engagement between the groups, although robust correlations between MST metrics and performance indices were lacking.

Online multiclass EEG feature extraction and recognition using modified convolutional neural network method

<p>Many techniques have been introduced to improve both brain-computer interface (BCI) steps: feature extraction and classification. One of the emerging trends in this field is the implementation of deep learning algorithms. There is a limited number of studies that investigated the application of deep learning techniques in electroencephalography (EEG) feature extraction and classification. This work is intended to apply deep learning for both stages: feature extraction and classification. This paper proposes a modified convolutional neural network (CNN) feature extractorclassifier algorithm to recognize four different EEG motor imagery (MI). In addition, a four-class linear discriminant analysis (LDR) classifier model was built and compared to the proposed CNN model. The paper showed very good results with 92.8% accuracy for one EEG four-class MI set and 85.7% for another set. The results showed that the proposed CNN model outperforms multi-class linear discriminant analysis with an accuracy increase of 28.6% and 17.9% for both MI sets, respectively. Moreover, it has been shown that majority voting for five repetitions introduced an accuracy advantage of 15% and 17.2% for both EEG sets, compared with single trials. This confirms that increasing the number of trials for the same MI gesture improves the recognition accuracy</p>

Diagnostic value of 2017 ACR Thyroid Imaging Reporting and Data System (TI-RADS) combined with fine needle aspiration biopsy in thyroid nodules

OBJECTIVE: To evaluate diagnostic value of Thyroid Imaging Reporting and Data System published by American College of Radiology (ACR TI-RADS) in 2017, ultrasound-guided fine-needle aspiration (US-FNA), and the combination of both methods in differentiation between benign and malignant thyroid nodules. METHODS: The data of US-FNA and ACR TI-RADS are collected from 159 patients underwent thyroid surgery in our hospital, which include a total of 178 thyroid nodules. A Bethesda System for Reporting Thyroid Cytopathology category of ≥IV and an ACR TI-RADS category ≥4 are regarded as diagnosis standards for malignancy in US-FNA and ACR TI-RADS, respectively. The pathological results after surgery are considered as the gold standard. The sensitivity, specificity, accuracy, positive predictive value and negative predictive value of the ACR TI-RADS, US-FNA and the combination of both methods for the differential diagnosis of thyroid nodules are calculated, respectively. RESULTS: The sensitivity, specificity and accuracy of ACR TI-RADS are 85.4%, 37.5%and 72.5%, respectively. The sensitivity, specificity and accuracy of US-FNA are 70.0%, 100%and 78.1%, respectively. After combining these two methods, the sensitivity, specificity and accuracy increase to 99.23%, 37.50%and 82.58%, respectively. The sensitivity of ACR TI-RADS is higher than that of US-FAN, and the sensitivity of combining these two methods is also higher than that of using ACR TI-RADS and US-FNA alone. CONCLUSION: The established ACR TI-RADS can help in selecting the target during nodule puncture, while the combination of ACR TI-RADS and US-FAN can further improve diagnostic ability for detecting malignant thyroid nodules.

Sentiment Classification Algorithm Based on the Cascade of BERT Model and Adaptive Sentiment Dictionary

The mobile social network contains a large amount of information in a form of commentary. Effective analysis of the sentiment in the comments would help improve the recommendations in the mobile network. With the development of well-performing pretrained language models, the performance of sentiment classification task based on deep learning has seen new breakthroughs in the past decade. However, deep learning models suffer from poor interpretability, making it difficult to integrate sentiment knowledge into the model. This paper proposes a sentiment classification model based on the cascade of the BERT model and the adaptive sentiment dictionary. First, the pretrained BERT model is used to fine-tune with the training corpus, and the probability of sentiment classification in different categories is obtained through the softmax layer. Next, to allow a more effective comparison between the probabilities for the two classes, a nonlinearity is introduced in a form of positive-negative probability ratio, using the rule method based on sentiment dictionary to deal with the probability ratio below the threshold. This method of cascading the pretrained model and the semantic rules of the sentiment dictionary allows to utilize the advantages of both models. Different sized Chnsenticorp data sets are used to train the proposed model. Experimental results show that the Dict-BERT model is better than the BERT-only model, especially when the training set is relatively small. The improvement is obvious with the accuracy increase of 0.8%.

Inter-observer variability between readers of CT images: all for one and one for all

BACKGROUND: The markup of medical image datasets is based on the subjective interpretation of the observed entities by radiologists. There is currently no widely accepted protocol for determining ground truth based on radiologists reports. AIM: To assess the accuracy of radiologist interpretations and their agreement for the publicly available dataset CTLungCa-500, as well as the relationship between these parameters and the number of independent readers of CT scans. MATERIALS AND METHODS: Thirty-four radiologists took part in the dataset markup. The dataset included 536 patients who were at high risk of developing lung cancer. For each scan, six radiologists worked independently to create a report. After that, an arbitrator reviewed the lesions discovered by them. The number of true-positive, false-positive, true-negative, and false-negative findings was calculated for each reader to assess diagnostic accuracy. Further, the inter-observer variability was analyzed using the percentage agreement metric. RESULTS: An increase in the number of independent readers providing CT scan interpretations leads to accuracy increase associated with a decrease in agreement. The majority of disagreements were associated with the presence of a lung nodule in a specific site of the CT scan. CONCLUSION: If arbitration is provided, an increase in the number of independent initial readers can improve their combined accuracy. The experience and diagnostic accuracy of individual readers have no bearing on the quality of a crowd-tagging annotation. At four independent readings per CT scan, the optimal balance of markup accuracy and cost was achieved.

Simulation of thermal behavior of nanoparticles due to appearance of MHD

To illustrate the role of Lorentz force on migration of nanopowders, CVFEM simulation has been reported in current research. The chamber contains hybrid nanomaterial and made up form porous media. Momentum equations have been modified for present paper with adding new source terms. The mentioned method works based on FEM in generation of mesh and calculation of gradient of scalars while it uses FVM approach for employing source terms. Testing with benchmark article shows the nice accuracy. Increase of permeability can enhance the speed of nanopowders and iso-temperature lines shapes become complicated. Impose of MHD creates new force against buoyancy and declines the velocity of the nanomaterial. Also, complication of isotherms declines with rise of Ha. With growth of Da, value of [Formula: see text] increases about 111% and 64.2% when [Formula: see text] and 20, respectively. Also, augment of Ha results in reduction of velocity about 30% and 47.6% when [Formula: see text] and 100. Given [Formula: see text], Nu for [Formula: see text] is 6.83 times bigger than case with [Formula: see text]. Nu decreases to about 67.28% with increase of Ha when [Formula: see text], [Formula: see text]. As Da increases, Nu rises about 62% when [Formula: see text], [Formula: see text].

Implementation of Rapid CMR Protocol in the COVID-19 era: improving scanning efficiency and increasing scanning capacity

Funding Acknowledgements Type of funding sources: None. Background During the COVID-19 pandemic, many non-urgent elective cardiac MRI (CMR) appointments were cancelled to minimise the risk of infection to patients coming to hospital. At the time of the first lockdown, our scanning schedule allowed on average 228 scans/month. Non-urgent elective studies were cancelled from April-June 2020, resulting in 684 scans added to the waiting list. Upon reactivation of our clinical CMR service, we developed a service quality improvement initiative consisting of using a ‘Rapid CMR’ protocol to reduce scanning time without compromising the test’s diagnostic accuracy, increase our scanning capacity and improve efficiency in reducing the backlog of requests. Purpose To demonstrate the increased scanning capacity generated by the adoption of the "Rapid CMR" protocol. Methods The Rapid CMR protocol was implemented in November 2020 to all scans requiring cines, late gadolinium enhancement ± adenosine stress (non-stress and stress studies). The protocol was modelled on prior published experiences[1,2]. Patients who underwent these scans with additional imaging (e.g. T2-STIR imaging) were excluded. Data was collected from Nov 2020 to Jan 2021 and compared with the same time period the previous year when the standard protocol was used (cf. Image 1). Data collected included scan duration (time from first to last image), whether the Rapid CMR studies maintained diagnostic quality (yes/no), and the did-not-attend (DNA) rate. Results With the Rapid CMR protocol 254 patients were scanned (114 non-stress, 140 stress), compared with 286 patients scanned with standard protocol in November 2019 to January 2020 (155 non-stress, 131 stress). Median scanning time in minutes for non-stress was 29 (IQR 25-34; Rapid) vs 37 (IQR 33-41; standard); (p &lt; 0.001). For stress studies the median scanning time in minutes was 32 (IQR 28-36; Rapid) vs 41 (IQR 29-45; standard; (p &lt; 0.001). The rate of suboptimal imaging due to patient factors (such as breathing or arrhythmia) was similar for each protocol (14.4% Rapid, 20.2% standard; p = 0.04). All Rapid studies were of diagnostic quality (Table 1). Saving c.8 minutes per scan led to an improved scanning time and schedule capacity of 21%. Fewer patients were scanned with the Rapid protocol due to pandemic related issues: patient reluctance to accept appointments (unfilled slots), cleaning measures between patients (on average ∼5 mins per slot reducing overall capacity), and a higher DNA rate: 15.3% (Rapid) vs 6.5% (standard); p &lt; 0.001. Conclusion The Rapid CMR protocol resulted in a statistically significant reduction in scanning time (-8 min for both stress and non-stress CMRs) increasing our schedule capacity and improving efficiency by 21%, whilst maintaining diagnostic quality. The implementation of the Rapid CMR protocol is a feasible and effective strategy to tackle the backlog of CMR clinical request accumulated during the pandemic.