Video based hand gesture recognition dataset using thermal camera

The dataset includes thermal videos of various hand gestures captured by the FLIR Lepton Thermal Camera. A large dataset is created to accurately classify hand gestures captured from eleven different individuals. The dataset consists of 9 classes corresponding to various hand gestures from different people collected at different time instances with complex backgrounds. This data includes flat/leftward, flat/rightward, flat/contract, spread/ leftward, spread/rightward, spread/contract, V-shape/leftward, V-shape/rightward, and V-shape/contract. There are 110 videos in the dataset for each gesture and a total of 990 videos corresponding to 9 gestures. Each video has data of three different (15 / 10 / 5 ) frame lengths.

SWCNTs/PEDOT:PSS nanocomposites-modified microelectrode arrays for revealing locking relations between burst and local field potential in cultured cortical networks

Burst and local field potential (LFP) are fundamental components of brain activity, representing fast and slow rhythms, respectively. Understanding the intricate relationship between burst and LFP is crucial for deciphering the underlying mechanisms of brain dynamics. In this study, we fabricated high-performance microelectrode arrays (MEAs) using the SWCNTs/PEDOT:PSS nanocomposites, which exhibited favorable electrical properties (low impedance: 12.8 ± 2.44 kΩ) and minimal phase delay (-11.96 ± 1.64°). These MEAs enabled precise exploration of the burst-LFP interaction in cultured cortical networks. After a 14-day period of culture, we used the MEAs to monitor electrophysiological activities and revealed a time-locking relationship between burst and LFP, indicating the maturation of the neural network. To further investigate this relationship, we modulated burst firing patterns by treating the neural culture with increasing concentrations of glycine. The results indicated that glycine effectively altered burst firing patterns, with both duration and spike count increasing as the concentration rose. This was accompanied by an enhanced level of time-locking between burst and LFP but a decrease in synchrony among neurons. This study not only highlighted the pivotal role of SWCNTs/PEDOT:PSS-modified MEAs in elucidating the interaction between burst and LFP, bridging the gap between slow and fast brain rhythms in vitro but also provides valuable insights into the potential therapeutic strategies targeting neurological disorders associated with abnormal rhythm generation.

Neural network establishes co-occurrence links between transformation products of the contaminant and the soil microbiome

It remains challenging to establish reliable links between transformation products (TPs) of contaminants and corresponding microbes. This challenge arises due to the sophisticated experimental regime required for TP discovery and the compositional nature of 16S rRNA gene amplicon sequencing and mass spectrometry datasets, which can potentially confound statistical inference. In this study, we present a new strategy by combining the use of 2H-labeled Stable Isotope-Assisted Metabolomics (2H-SIAM) with a neural network-based algorithm (i.e., MMvec) to explore links between TPs of pyrene and the soil microbiome. The links established by this novel strategy were further validated using different approaches. Briefly, a metagenomic study provided indirect evidence for the established links, while the identification of pyrene degraders from soils, and a DNA-based stable isotope probing (DNA-SIP) study offered direct evidence. The comparison among different approaches, including Pearson's and Spearman's correlations, further confirmed the superior performance of our strategy. In conclusion, we summarize the unique features of the combined use of 2H-SIAM and MMvec. This study not only addresses the challenges in linking TPs to microbes but also introduces an innovative and effective approach for such investigations. Environmental Implication: Taxonomically diverse bacteria performing successive metabolic steps of the contaminant were firstly depicted in the environmental matrix.

A novel feature-level fusion scheme with multimodal attention CNN for heart sound classification

BACKGROUND AND OBJECTIVE

Most of the existing machine learning-based heart sound classification methods achieve limited accuracy. Since they primarily depend on single domain feature information and tend to focus equally on each part of the signal rather than employing a selective attention mechanism. In addition, they fail to exploit convolutional neural network (CNN) - based features with an effective fusion strategy.

METHODS

In order to overcome these limitations, a novel multimodal attention convolutional neural network (MACNN) with a feature-level fusion strategy, in which Mel-cepstral domain as well as general frequency domain features are incorporated to increase the diversity of the features, is proposed in this paper. In the proposed method, DilationAttenNet is first utilized to construct attention-based CNN feature extractors and then these feature extractors are jointly optimized in MACNN at the feature-level. The attention mechanism aims to suppress irrelevant information and focus on crucial diverse features extracted from the CNN.

RESULTS

Extensive experiments are carried out to study the efficacy of the feature level fusion in comparison to that with early fusion. The results show that the proposed MACNN method significantly outperforms the state-of-the-art approaches in terms of accuracy and score for the two publicly available Github and Physionet datasets.

CONCLUSION

The findings of our experiments demonstrated the high performance for heart sound classification based on the proposed MACNN, and hence have potential clinical usefulness in the identification of heart diseases. This technique can assist cardiologists and researchers in the design and development of heart sound classification methods.

Data-driven learning of chaotic dynamical systems using Discrete-Temporal Sobolev Networks

We introduce the Discrete-Temporal Sobolev Network (DTSN), a neural network loss function that assists dynamical system forecasting by minimizing variational differences between the network output and the training data via a temporal Sobolev norm. This approach is entirely data-driven, architecture agnostic, and does not require derivative information from the estimated system. The DTSN is particularly well suited to chaotic dynamical systems as it minimizes noise in the network output which is crucial for such sensitive systems. For our test cases we consider discrete approximations of the Lorenz-63 system and the Chua circuit. For the network architectures we use the Long Short-Term Memory (LSTM) and the Transformer. The performance of the DTSN is compared with the standard MSE loss for both architectures, as well as with the Physics Informed Neural Network (PINN) loss for the LSTM. The DTSN loss is shown to substantially improve accuracy for both architectures, while requiring less information than the PINN and without noticeably increasing computational time, thereby demonstrating its potential to improve neural network forecasting of dynamical systems.

Enhancing and improving the performance of imbalanced class data using novel GBO and SSG: A comparative analysis

Class imbalance problem (CIP) in a dataset is a major challenge that significantly affects the performance of Machine Learning (ML) models resulting in biased predictions. Numerous techniques have been proposed to address CIP, including, but not limited to, Oversampling, Undersampling, and cost-sensitive approaches. Due to its ability to generate synthetic data, oversampling techniques such as the Synthetic Minority Oversampling Technique (SMOTE) are the most widely used methodology by researchers. However, one of SMOTE's potential disadvantages is that newly created minor samples overlap with major samples. Therefore, the probability of ML models' biased performance toward major classes increases. Generative adversarial network (GAN) has recently garnered much attention due to their ability to create real samples. However, GAN is hard to train even though it has much potential. Considering these opportunities, this work proposes two novel techniques: GAN-based Oversampling (GBO) and Support Vector Machine-SMOTE-GAN (SSG) to overcome the limitations of the existing approaches. The preliminary results show that SSG and GBO performed better on the nine imbalanced benchmark datasets than several existing SMOTE-based approaches. Additionally, it can be observed that the proposed SSG and GBO methods can accurately classify the minor class with more than 90% accuracy when tested with 20%, 30%, and 40% of the test data. The study also revealed that the minor sample generated by SSG demonstrates Gaussian distributions, which is often difficult to achieve using original SMOTE and SVM-SMOTE.

A machine learning based method for constructing group profiles of university students

This study facilitates university student profiling by constructing a prediction model to forecast the classification of future students participating in a survey, thereby enhancing the utility and effectiveness of the questionnaire approach. In the context of the ongoing digital transformation of campuses, higher education institutions are increasingly prioritizing student educational development. This shift aligns with the maturation of big data technology, prompting scholars to focus on profiling university student education. While earlier research in this area, particularly foreign studies, focus on extracting data from specific learning contexts and often relied on single data sources, our study addresses these limitations. We employ a comprehensive approach, incorporating questionnaire surveys to capture a diverse array of student data. Considering various university student attributes, we create a holistic profile of the student population. Furthermore, we use clustering techniques to develop a categorical prediction model. In our clustering analysis, we employ the K-means algorithm to group student survey data. The results reveal four distinct student profiles: Diligent Learners, Earnest Individuals, Discerning Achievers, and Moral Advocates. These profiles are subsequently used to label student groups. For the classification task, we leverage these labels to establish a prediction model based on the Back Propagation neural network, with the goal of assigning students to their respective groups. Through meticulous model optimization, an impressive classification accuracy of 90.22% is achieved. Our research offers a novel perspective and serves as a valuable methodological reference for university student profiling.

The application of improved densenet algorithm in accurate image recognition

Image recognition technology belongs to an important research field of artificial intelligence. In order to enhance the application value of image recognition technology in the field of computer vision and improve the technical dilemma of image recognition, the research improves the feature reuse method of dense convolutional network. Based on gradient quantization, traditional parallel algorithms have been improved. This improvement allows for independent parameter updates layer by layer, reducing communication time and data volume. The introduction of quantization error reduces the impact of gradient loss on model convergence. The test results show that the improvement strategy designed by the research improves the model parameter efficiency while ensuring the recognition effect. Narrowing the learning rate is conducive to refining the updating granularity of model parameters, and deepening the number of network layers can effectively improve the final recognition accuracy and convergence effect of the model. It is better than the existing state-of-the-art image recognition models, visual geometry group and EfficientNet. The parallel acceleration algorithm, which is improved by the gradient quantization, performs better than the traditional synchronous data parallel algorithm, and the improvement of the acceleration ratio is obvious. Compared with the traditional synchronous data parallel algorithm and stale synchronous parallel algorithm, the optimized parallel acceleration algorithm of the study ensures the image data training speed and solves the bottleneck problem of communication data. The model designed by the research improves the accuracy and training speed of image recognition technology and expands the use of image recognition technology in the field of computer vision.Please confirm the affiliation details of [1] is correct.The relevant detailed information in reference [1] has been confirmed to be correct.

Intelligent salivary biosensors for periodontitis: in vitro simulation of oral oxidative stress conditions

ASTRACT

One of the most common oral diseases affecting millions of people worldwide is periodontitis. Usually, proteins in body fluids are used as biomarkers of diseases. This study focused on hydrogen peroxide, lipopolysaccharide (LPS), and lactic acid as salivary non-protein biomarkers for oxidative stress conditions of periodontitis. Electrochemical analysis of artificial saliva was done using Gamry with increasing hydrogen peroxide, bLPS, and lactic acid concentrations. Electrochemical impedance spectroscopy (EIS) and cyclic voltammetry (CV) were conducted. From EIS data, change in capacitance and CV plot area were calculated for each test condition. Hydrogen peroxide groups had a decrease in CV area and an increase in percentage change in capacitance, lipopolysaccharide groups had a decrease in CV area and a decrease in percentage change in capacitance, and lactic acid groups had an increase of CV area and an increase in percentage change in capacitance with increasing concentrations. These data showed a unique combination of electrochemical properties for the three biomarkers. Scanning electron microscopy (SEM) with energy dispersive spectroscopy (EDS) employed to observe the change in the electrode surface and elemental composition data present on the sensor surface also showed a unique trend of elemental weight percentages. Machine learning models using hydrogen peroxide, LPS, and lactic acid electrochemical data were applied for the prediction of risk levels of periodontitis. The detection of hydrogen peroxide, LPS, and lactic acid by electrochemical biosensors indicates the potential to use these molecules as electrochemical biomarkers and use the data for ML-driven prediction tool for the periodontitis risk levels.

Existence of reservoir with finite-dimensional output for universal reservoir computing

In this paper, we prove the existence of a reservoir that has a finite-dimensional output and makes the reservoir computing model universal. Reservoir computing is a method for dynamical system approximation that trains the static part of a model but fixes the dynamical part called the reservoir. Hence, reservoir computing has the advantage of training models with a low computational cost. Moreover, fixed reservoirs can be implemented as physical systems. Such reservoirs have attracted attention in terms of computation speed and energy consumption. The universality of a reservoir computing model is its ability to approximate an arbitrary system with arbitrary accuracy. Two sufficient reservoir conditions to make the model universal have been proposed. The first is the combination of fading memory and the separation property. The second is the neighborhood separation property, which we proposed recently. To date, it has been unknown whether a reservoir with a finite-dimensional output can satisfy these conditions. In this study, we prove that no reservoir with a finite-dimensional output satisfies the former condition. By contrast, we propose a single output reservoir that satisfies the latter condition. This implies that, for any dimension, a reservoir making the model universal exists with the output of that specified dimension. These results clarify the practical importance of our proposed conditions.

Satin bowerbird optimizer-neural network for approximating the capacity of CFST columns under compression

Concrete-filled steel tube columns (CFSTCs) are important elements in the construction sector and predictive analysis of their behavior is essential. Recent works have revealed the potential of metaheuristic-assisted approximators for this purpose. The main idea of this paper, therefore, is to introduce a novel integrative model for appraising the axial compression capacity (Pu) of CFSTCs. The proposed model represents an artificial neural network (ANN) supervised by satin bowerbird optimizer (SBO). In other words, this metaheuristic algorithm trains the ANN optimally to find the best contribution of input parameters to the Pu. In this sense, column length and the compressive strength of concrete, as well as the characteristics of the steel tube (i.e., diameter, thickness, yield stress, and ultimate stress), are considered input data. The prediction results are compared to five ANNs supervised by backtracking search algorithm (BSA), earthworm optimization algorithm (EWA), social spider algorithm (SOSA), salp swarm algorithm (SSA), and wind-driven optimization. Evaluating various accuracy indicators showed that the proposed model surpassed all of them in both learning and reproducing the Pu pattern. The obtained values of mean absolute percentage error of the SBO-ANN was 2.3082% versus 4.3821%, 17.4724%, 15.7898%, 4.2317%, and 3.6884% for the BSA-ANN, EWA-ANN, SOSA-ANN, SSA-ANN and WDA-ANN, respectively. The higher accuracy of the SBO-ANN against several hybrid models from earlier literature was also deduced. Moreover, the outcomes of principal component analysis on the dataset showed that the yield stress, diameter, and ultimate stress of the steel tube are the three most important factors in Pu prediction. A predictive formula is finally derived from the optimized SBO-ANN by extracting and organizing the weights and biases of the ANN. Owing to the accurate estimation shown by this model, the derived formula can reliably predict the Pu of concrete-filled steel tube columns.

Game-theoretic optimization of landslide susceptibility mapping: a comparative study between Bayesian-optimized basic neural network and new generation neural network models

Landslide susceptibility mapping is essential for reducing the risk of landslides and ensuring the safety of people and infrastructure in landslide-prone areas. However, little research has been done on the development of well-optimized Elman neural networks (ENN), deep neural networks (DNN), and artificial neural networks (ANN) for robust landslide susceptibility mapping (LSM). Additionally, there is a research gap regarding the use of Bayesian optimization and the derivation of SHapley Additive exPlanations (SHAP) values from optimized models. Therefore, this study aims to optimize DNN, ENN, and ANN models using Bayesian optimization for landslide susceptibility mapping and derive SHAP values from these optimized models. The LSM models have been validated using the receiver operating characteristics curve, confusion matrix, and other twelve error matrices. The study used six machine learning-based feature selection techniques to identify the most important variables for predicting landslide susceptibility. The decision tree, random forest, and bagging feature selection models showed that slope, elevation, DFR, annual rainfall, LD, DD, RD, and LULC are influential variables, while geology and soil texture have less influence. The DNN model outperformed the other two models, covering 7839.54 km2 under the very low landslide susceptibility zone and 3613.44 km2 under the very high landslide susceptibility zone. The DNN model is better suited for generating landslide susceptibility maps, as it can classify areas with higher accuracy. The model identified several key factors that contribute to the initiation of landslides, including high elevation, built-up and agricultural land use, less vegetation, aspect (north and northwest), soil depth less than 140 cm, high rainfall, high lineament density, and a low distance from roads. The study's findings can help stakeholders make informed decisions to reduce the risk of landslides and ensure the safety of people and infrastructure in landslide-prone areas.

Carbon emission prediction models: A review

Amidst growing concerns over the greenhouse effect, especially its consequential impacts, establishing effective Carbon Emission Prediction Models (CEPMs) to comprehend and predict CO2 emission trends is imperative for climate change mitigation. A review of 147 Carbon Emission Prediction Model (CEPM) studies revealed three predominant functions-prediction, optimization, and prediction factor selection. Statistical models, comprising 75 instances, were the most prevalent among prediction models, followed by neural network models at 21.8 %. The consistent rise in neural network model usage, particularly feedforward architectures, was observed from 2019 to 2022. A majority of CEPMs incorporated optimized approaches, with 94.4 % utilizing metaheuristic models. Parameter optimization was the primary focus, followed by structure optimization. Prediction factor selection models, employing Grey Relational Analysis (GRA) and Principal Component Analysis (PCA) for statistical and machine learning models, respectively, filtered factors effectively. Scrutinizing accuracy, pre-optimized CEPMs exhibited varied performance, Root Mean Square Error (RMSE) values spanned from 0.112 to 1635 Mt, while post-optimization led to a notable improvement, the minimum RMSE reached 0.0003 Mt, and the maximum was 95.14 Mt. Finally, we summarized the pros and cons of existing models, classified and counted the factors that influence carbon emissions, clarified the research objectives in CEPM and assessed the applied model evaluation methods and the spatial and temporal scales of existing research.

Prediction of adverse cardiovascular events in children using artificial intelligence-based electrocardiogram

BACKGROUND

Convolutional neural networks (CNNs) have emerged as a novel method for evaluating heart failure (HF) in adult electrocardiograms (ECGs). However, such CNNs are not applicable to pediatric HF, where abnormal anatomy of congenital heart defects plays an important role. ECG-based CNNs reflecting neurohormonal activation (NHA) may be a useful marker of pediatric HF. This study aimed to develop and validate an ECG-derived marker of pediatric HF that reflects the risk of future cardiovascular events.

METHODS

Based on 21,378 ECGs from 8324 children, a CNN was trained using B-type natriuretic peptide (BNP) and the occurrence of major adverse cardiovascular events (MACEs). The output of the model, or the electrical heart failure indicator (EHFI), was compared with the BNP regarding its ability to predict MACEs in 813 ECGs from 295 children.

RESULTS

EHFI achieved a better area under the curve than BNP in predicting MACEs within 180 days (0.826 versus 0.691, p = 0.03). On Cox univariable analyses, both EHFI and BNP were significantly associated with MACE (log10 EHFI: hazard ratio [HR] = 16.5, p < 0.005 and log10 BNP: HR = 4.4, p < 0.005). The time-dependent average precisions of EHFI in predicting MACEs were 32.4%-67.9% and 1.6-7.5-fold higher than those of BNP in the early period. Additionally, the MACE rate increased monotonically with EHFI, whereas the rate peaked at approximately 100 pg/mL of BNP and decreased in the higher range.

CONCLUSIONS

ECG-derived CNN is a novel marker of HF with different prognostic potential from BNP. CNN-based ECG analysis may provide a new guide for assessing pediatric HF.

Deep learning-based computer-aided diagnosis system for the automatic detection and classification of lateral cervical lymph nodes on original ultrasound images of papillary thyroid carcinoma: a prospective diagnostic study

PURPOSE

This study aims to develop a deep learning-based computer-aided diagnosis (CAD) system for the automatic detection and classification of lateral cervical lymph nodes (LNs) on original ultrasound images of papillary thyroid carcinoma (PTC) patients.

METHODS

A retrospective data set of 1801 cervical LN ultrasound images from 1675 patients with PTC and a prospective test set including 185 images from 160 patients were collected. Four different deep leaning models were trained and validated in the retrospective data set. The best model was selected for CAD system development and compared with three sonographers in the retrospective and prospective test sets.

RESULTS

The Deformable Detection Transformer (DETR) model showed the highest diagnostic efficacy, with a mean average precision score of 86.3% in the retrospective test set, and was therefore used in constructing the CAD system. The detection performance of the CAD system was superior to the junior sonographer and intermediate sonographer with accuracies of 86.3% and 92.4% in the retrospective and prospective test sets, respectively. The classification performance of the CAD system was better than all sonographers with the areas under the curve (AUCs) of 94.4% and 95.2% in the retrospective and prospective test sets, respectively.

CONCLUSIONS

This study developed a Deformable DETR model-based CAD system for automatically detecting and classifying lateral cervical LNs on original ultrasound images, which showed excellent diagnostic efficacy and clinical utility. It can be an important tool for assisting sonographers in the diagnosis process.

Computer Vision for Identification of Increased Fetal Heart Variability in Cardiotocogram

INTRODUCTION

Increased fetal heart rate variability (IFHRV), defined as fetal heart rate (FHR) baseline amplitude changes of >25 beats per minute with a duration of ≥1 min, is an early sign of intrapartum fetal hypoxia. This study evaluated the level of agreement of machine learning (ML) algorithms-based recognition of IFHRV patterns with expert analysis.

METHODS

Cardiotocographic recordings and cardiotocograms from 4,988 singleton term childbirths were evaluated independently by two expert obstetricians blinded to the outcomes. Continuous FHR monitoring with computer vision analysis was compared with visual analysis by the expert obstetricians. FHR signals were graphically processed and measured by the computer vision model labeled SALKA.

RESULTS

In visual analysis, IFHRV pattern occurred in 582 cardiotocograms (11.7%). Compared with visual analysis, SALKA recognized IFHRV patterns with an average Cohen's kappa coefficient of 0.981 (95% CI: 0.972-0.993). The sensitivity of SALKA was 0.981, the positive predictive rate was 0.822 (95% CI: 0.774-0.903), and the false-negative rate was 0.01 (95% CI: 0.00-0.02). The agreement between visual analysis and SALKA in identification of IFHRV was almost perfect (0.993) in cases (N = 146) with neonatal acidemia (i.e., umbilical artery pH <7.10).

CONCLUSIONS

Computer vision analysis by SALKA is a novel ML technique that, with high sensitivity and specificity, identifies IFHRV features in intrapartum cardiotocograms. SALKA recognizes potential early signs of fetal distress close to those of expert obstetricians, particularly in cases of neonatal acidemia.

Cholestatic liver disease leads to significant adaptative changes in neural circuits regulating social behavior in mice to enhance sociability

BACKGROUND & AIMS

Cholestatic liver diseases (CLD) are commonly associated with behavioral changes, including social isolation, that negatively affects patient quality of life and remains unaltered by current therapies. It remains unclear whether CLD-associated social dysfunction stems from a direct effect on the brain, or from the psychological impact of CLD. The psychological component of disease is absent in animals, so we investigated the impact of CLD on social behavior and gene expression profiles in key social behavior-regulating brain regions in a mouse model.

METHODS

CLD due to bile duct ligation was used with the three-chamber sociability test for behavioral phenotyping. Differentially expressed gene (DEG) signatures were delineated in 3 key brain regions regulating social behavior using RNA-seq. Ingenuity Pathway Analysis (IPA®) was applied to streamline DEG data interpretation and integrate findings with social behavior-regulating pathways to identify important brain molecular networks and regulatory mechanisms disrupted in CLD.

RESULTS

CLD mice exhibited enhanced social interactive behavior and significantly altered gene expression in each of the three social behavior-regulating brain regions examined. DEG signatures in BDL mice were associated with key IPA®-identified social behavior-regulating pathways including Oxytocin in Brain Signaling, GABA Receptor Signaling, Dopamine Receptor Signaling, and Glutamate Receptor Signaling.

CONCLUSIONS

CLD causes complex alterations in gene expression profiles in key social behavior-regulating brain areas/pathways linked to enhanced social interactive behavior. These findings, if paralleled in CLD patients, suggest that CLD-associated reductions in social interactions predominantly relate to psychological impacts of disease and may inform new approaches to improve management.

A new cloud-based method for composition of healthcare services using deep reinforcement learning and Kalman filtering

Healthcare has significantly contributed to the well-being of individuals around the globe; nevertheless, further benefits could be derived from a more streamlined healthcare system without incurring additional costs. Recently, the main attributes of cloud computing, such as on-demand service, high scalability, and virtualization, have brought many benefits across many areas, especially in medical services. It is considered an important element in healthcare services, enhancing the performance and efficacy of the services. The current state of the healthcare industry requires the supply of healthcare products and services, increasing its viability for everyone involved. Developing new approaches for discovering and selecting healthcare services in the cloud has become more critical due to the rising popularity of these kinds of services. As a result of the diverse array of healthcare services, service composition enables the execution of intricate operations by integrating multiple services' functionalities into a single procedure. However, many methods in this field encounter several issues, such as high energy consumption, cost, and response time. This article introduces a novel layered method for selecting and evaluating healthcare services to find optimal service selection and composition solutions based on Deep Reinforcement Learning (Deep RL), Kalman filtering, and repeated training, addressing the aforementioned issues. The results revealed that the proposed method has achieved acceptable results in terms of availability, reliability, energy consumption, and response time when compared to other methods.

AI-based automated detection and stability analysis of traumatic vertebral body fractures on computed tomography

PURPOSE

We developed and tested a neural network for automated detection and stability analysis of vertebral body fractures on computed tomography (CT).

MATERIALS AND METHODS

257 patients who underwent CT were included in this Institutional Review Board (IRB) approved study. 463 fractured and 1883 non-fractured vertebral bodies were included, with 190 fractures unstable. Two readers identified vertebral body fractures and assessed their stability. A combination of a Hierarchical Convolutional Neural Network (hNet) and a fracture Classification Network (fNet) was used to build a neural network for the automated detection and stability analysis of vertebral body fractures on CT. Two final test settings were chosen: one with vertebral body levels C1/2 included and one where they were excluded.

RESULTS

The mean age of the patients was 68 ± 14 years. 140 patients were female. The network showed a slightly higher diagnostic performance when excluding C1/2. Accordingly, the network was able to distinguish fractured and non-fractured vertebral bodies with a sensitivity of 75.8 % and a specificity of 80.3 %. Additionally, the network determined the stability of the vertebral bodies with a sensitivity of 88.4 % and a specificity of 80.3 %. The AUC was 87 % and 91 % for fracture detection and stability analysis, respectively. The sensitivity of our network in indicating the presence of at least one fracture / one unstable fracture within the whole spine achieved values of 78.7 % and 97.2 %, respectively, when excluding C1/2.

CONCLUSION

The developed neural network can automatically detect vertebral body fractures and evaluate their stability concurrently with a high diagnostic performance.

Automatic Smoke Analysis in Minimally Invasive Surgery by Image-Based Machine Learning

INTRODUCTION

Minimally Invasive Surgery uses electrosurgical tools that generate smoke. This smoke reduces the visibility of the surgical site and spreads harmful substances with potential hazards for the surgical staff. Automatic image analysis may provide assistance. However, the existing studies are restricted to simple clear versus smoky image classification.

MATERIALS AND METHODS

We propose a novel approach using surgical image analysis with machine learning, including deep neural networks. We address three tasks: 1) smoke quantification, which estimates the visual level of smoke, 2) smoke evacuation confidence, which estimates the level of confidence to evacuate smoke, and 3) smoke evacuation recommendation, which estimates the evacuation decision. We collected three datasets with expert annotations. We trained end-to-end neural networks for the three tasks. We also created indirect predictors using task 1 followed by linear regression to solve task 2 and using task 2 followed by binary classification to solve task 3.

RESULTS

We observe a reasonable inter-expert variability for tasks 1 and a large one for tasks 2 and 3. For task 1, the expert error is 17.61 percentage points (pp) and the neural network error is 18.45 pp. For tasks 2, the best results are obtained from the indirect predictor based on task 1. For this task, the expert error is 27.35 pp and the predictor error is 23.60 pp. For task 3, the expert accuracy is 76.78% and the predictor accuracy is 81.30%.

CONCLUSIONS

Smoke quantification, evacuation confidence, and evaluation recommendation can be achieved by automatic surgical image analysis with similar or better accuracy as the experts.

Circuits in the motor cortex explain oscillatory responses to transcranial magnetic stimulation

Transcranial magnetic stimulation (TMS) is a popular method used to investigate brain function. Stimulation over the motor cortex evokes muscle contractions known as motor evoked potentials (MEPs) and also high-frequency volleys of electrical activity measured in the cervical spinal cord. The physiological mechanisms of these experimentally derived responses remain unclear, but it is thought that the connections between circuits of excitatory and inhibitory neurons play a vital role. Using a spiking neural network model of the motor cortex, we explained the generation of waves of activity, so called 'I-waves', following cortical stimulation. The model reproduces a number of experimentally known responses including direction of TMS, increased inhibition, and changes in strength. Using populations of thousands of neurons in a model of cortical circuitry we showed that the cortex generated transient oscillatory responses without any tuning, and that neuron parameters such as refractory period and delays influenced the pattern and timing of those oscillations. By comparing our network with simpler, previously proposed circuits, we explored the contributions of specific connections and found that recurrent inhibitory connections are vital in producing later waves that significantly impact the production of motor evoked potentials in downstream muscles (Thickbroom, 2011). This model builds on previous work to increase our understanding of how complex circuitry of the cortex is involved in the generation of I-waves.

Physics-guided neural network for predicting asphalt mixture rutting with balanced accuracy, stability and rationality

The prediction of rutting performance of asphalt materials poses a significant challenge due to the intricate relationships between the rutting performance and its influencing factors. Machine learning models have gained popularity to address this challenge by offering sophisticated model structures and algorithms. However, existing models often prioritize model accuracy over stability and rationality. The increasingly complicated model structure may lead to an imbalance between the data and the model, resulting in issues such as overfitting and reduced model applicability and interpretability. In this context, this study proposes a novel modeling framework to predict the rutting performance of asphalt mixture by utilizing autoencoder for feature selection and feedforward neural network for rut depth prediction. Notably, physics information of the selected variables is implemented into the neural network to achieve the appropriate balance of model accuracy, stability, and rationality. The results demonstrate that while maintaining high model accuracy, the implementation of physics information significantly enhances the model's stability and rationality. This framework holds great potential for accurate and reliable predictions of pavement distress by leveraging the complementary strengths of data-driven machine learning and physics-based domain knowledge.

Online biomedical named entities recognition by data and knowledge-driven model

Named entity recognition (NER) is an important task for the natural language processing of biomedical text. Currently, most NER studies standardized biomedical text, but NER for unstandardized biomedical text draws less attention from researchers. Named entities in online biomedical text exist with errors and polymorphisms, which negatively impact NER models' performance and impede support from knowledge representation methods. In this paper, we propose a neural network method that can effectively recognize entities in unstandardized online medical/health text. We introduce a new pre-training scheme that uses large-scale online question-answering pairs to enhance transformers' model capacity on online biomedical text. Moreover, we supply models with knowledge representations from a knowledge base called multi-channel knowledge labels, and this method overcomes the restriction from languages, like Chinese, that require word segmentation tools to represent knowledge. Our model outperforms other baseline methods significantly in experiments on a dataset for Chinese online medical entity recognition and achieves state-of-the-art results.

Comparison of modeling, optimization, and prediction of important parameters in the adsorption of cefixime onto sol-gel derived carbon aerogel and modified with nickel using ANN, RSM, GA, and SOLVER methods

Today, the main goal of many researchers is the use of high-performance, economically and industrially justified materials, as well as recyclable materials in removing organic and dangerous pollutants. For this purpose, sol-gel derived carbon aerogel modified with nickel (SGCAN) was used to remove Cefixime from aqueous solutions. The influence of important parameters in the cefixime adsorption onto SGCAN was modeled and optimized using artificial neural network (ANN), response surface methodology (RSM), genetic algorithm (GA), and SOLVER methods. R software was applied for this purpose. The design range of the runs for a time was in the range of 5 min-70 min, concentration in the range of 5 mg L-1 to 40 mg L-1, amount of adsorbent in the range of 0.05 g L-1 to 0.15 g L-1, and pH in the range of 2.0-11. The results showed that the ANN model due to lower Mean Squared Error (MSE), Sum of Squared Errors (SSE), and Root Mean Squared Error (RMSE) values and also higher R2 is a superior model than RSM. Also, due to the superiority of ANN over the RSM model, the optimum results were calculated based on GA. Based on GA, the highest Cefixime adsorption onto SGCAN was obtained in pH, 5.98; reaction time, 58.15 min; initial Cefixime concentration, 15.26 mg L-1; and adsorbent dosage, 0.11 g L-1. The maximum adsorption capacity of Cefixime onto SGCAN was determined to be 52 mg g-1. It was found the pseudo-second-order model has a better fit with the presented data.

Combined quantitative lipidomics and back-propagation neural network approach to discriminate the breed and part source of lamb

The study successfully utilized an analytical approach that combined quantitative lipidomics with back-propagation neural networks to identify breed and part source of lamb using small-scale samples. 1230 molecules across 29 lipid classes were identified in longissimus dorsi and knuckle meat of both Tan sheep and Bahan crossbreed sheep. Applying multivariate statistical methods, 12 and 7 lipid molecules were identified as potential markers for breed and part identification, respectively. Stepwise linear discriminant analysis was applied to select 3 and 4 lipid molecules, respectively, for discriminating lamb breed and part sources, achieving correct rates of discrimination of 100 % and 95 %. Additionally, back-propagation neural network proved to be a superior method for identifying sources of lamb meat compared to other machine learning approaches. These findings indicate that integrating lipidomics with back-propagation neural network approach can provide an effective strategy to trace and certify lamb products, ensuring their quality and protecting consumer rights.

Generalized latent multi-view clustering with tensorized bipartite graph

Tensor-based multi-view spectral clustering algorithms use tensors to model the structure of multi-dimensional data to take advantage of the complementary information and high-order correlations embedded in the graph, thus achieving impressive clustering performance. However, these algorithms use linear models to obtain consensus, which prevents the learned consensus from adequately representing the nonlinear structure of complex data. In order to address this issue, we propose a method called Generalized Latent Multi-View Clustering with Tensorized Bipartite Graph (GLMC-TBG). Specifically, in this paper we introduce neural networks to learn highly nonlinear mappings that encode nonlinear structures in graphs into latent representations. In addition, multiple views share the same latent consensus through nonlinear interactions. In this way, a more comprehensive common representation from multiple views can be achieved. An Augmented Lagrangian Multiplier with Alternating Direction Minimization (ALM-ADM) framework is designed to optimize the model. Experiments on seven real-world data sets verify that the proposed algorithm is superior to state-of-the-art algorithms.

Detection of lung cancer through SERS analysis of serum

Early detection and postoperative assessment are crucial for improving overall survival among lung cancer patients. Here, we report a non-invasive technique that integrates Raman spectroscopy with machine learning for the detection of lung cancer. The study encompassed 88 postoperative lung cancer patients, 73 non-surgical lung cancer patients, and 68 healthy subjects. The primary aim was to explore variations in serum metabolism across these cohorts. Comparative analysis of average Raman spectra was conducted, while principal component analysis was employed for data visualization. Subsequently, the augmented dataset was used to train convolutional neural networks (CNN) and Resnet models, leading to the development of a diagnostic framework. The CNN model exhibited superior performance, as verified by the receiver operating characteristic curve. Notably, postoperative patients demonstrated an increased likelihood of recurrence, emphasizing the crucial need for continuous postoperative monitoring. In summary, the integration of Raman spectroscopy with CNN-based classification shows potential for early detection and postoperative assessment of lung cancer.

Predicting quantum emitter fluctuations with time-series forecasting models

2D materials have important fundamental properties allowing for their use in many potential applications, including quantum computing. Various Van der Waals materials, including Tungsten disulfide (WS2), have been employed to showcase attractive device applications such as light emitting diodes, lasers and optical modulators. To maximize the utility and value of integrated quantum photonics, the wavelength, polarization and intensity of the photons from a quantum emission (QE) must be stable. However, random variation of emission energy, caused by the inhomogeneity in the local environment, is a major challenge for all solid-state single photon emitters. In this work, we assess the random nature of the quantum fluctuations, and we present time series forecasting deep learning models to analyse and predict QE fluctuations for the first time. Our trained models can roughly follow the actual trend of the data and, under certain data processing conditions, can predict peaks and dips of the fluctuations. The ability to anticipate these fluctuations will allow physicists to harness quantum fluctuation characteristics to develop novel scientific advances in quantum computing that will greatly benefit quantum technologies.

HairNet2: deep learning to quantify cotton leaf hairiness, a complex genetic and environmental trait

BACKGROUND

Cotton accounts for 80% of the global natural fibre production. Its leaf hairiness affects insect resistance, fibre yield, and economic value. However, this phenotype is still qualitatively assessed by visually attributing a Genotype Hairiness Score (GHS) to a leaf/plant, or by using the HairNet deep-learning model which also outputs a GHS. Here, we introduce HairNet2, a quantitative deep-learning model which detects leaf hairs (trichomes) from images and outputs a segmentation mask and a Leaf Trichome Score (LTS).

RESULTS

Trichomes of 1250 images were annotated (AnnCoT) and a combination of six Feature Extractor modules and five Segmentation modules were tested alongside a range of loss functions and data augmentation techniques. HairNet2 was further validated on the dataset used to build HairNet (CotLeaf-1), a similar dataset collected in two subsequent seasons (CotLeaf-2), and a dataset collected on two genetically diverse populations (CotLeaf-X). The main findings of this study are that (1) leaf number, environment and image position did not significantly affect results, (2) although GHS and LTS mostly correlated for individual GHS classes, results at the genotype level revealed a strong LTS heterogeneity within a given GHS class, (3) LTS correlated strongly with expert scoring of individual images.

CONCLUSIONS

HairNet2 is the first quantitative and scalable deep-learning model able to measure leaf hairiness. Results obtained with HairNet2 concur with the qualitative values used by breeders at both extremes of the scale (GHS 1-2, and 5-5+), but interestingly suggest a reordering of genotypes with intermediate values (GHS 3-4+). Finely ranking mild phenotypes is a difficult task for humans. In addition to providing assistance with this task, HairNet2 opens the door to selecting plants with specific leaf hairiness characteristics which may be associated with other beneficial traits to deliver better varieties.

Water-richness evaluation method and application of clastic rock aquifer in mining seam roof

Clastic rock aquifer of the coal seam roof often constitutes the direct water-filling aquifer of the coal seam and its water-richness is closely related to the risk of roof water inrush. Therefore, the evaluation of the water-richness of clastic rock aquifer is the basic work of coal seam roof water disaster prevention. This article took the 4th coal seam in Huafeng mine field as an example. It combined the empirical formula method and generalized regression neural network (GRNN) to calculate the development height of water-conducting fracture zone, determined the vertical spatial range of water-richness evaluation. Depth of the sandstone floor, brittle rock ratio, lithological structure index, fault strength index, and fault intersections and endpoints density were selected as the main controlling factors. A combination weighting method based on the analytic hierarchy process (AHP), rough set theory (RS), and minimum deviation method (MD) was proposed to determine the weight of the main controlling factors. Introduced the theory of unascertained measures and confidence recognition criteria to construct an evaluation model for the water-richness of clastic rock aquifers, the study area was divided into three zones: relatively weak water-richness zones, medium water-richness zones, and relatively strong water-richness zones. By comparing with the water inrush points and the water inflow of workfaces, the evaluation model's water yield zoning was consistent with the actual situation, and the prediction effect was good. This provided a new idea for the evaluation of the water-richness of the clastic rock aquifer on the roof of the mining coal seam.

Early identification of suspected serious infection among patients afebrile at initial presentation using neural network models and natural language processing: A development and external validation study in the emergency department

OBJECTIVE

To develop and externally validate models based on neural networks and natural language processing (NLP) to identify suspected serious infections in emergency department (ED) patients afebrile at initial presentation.

METHODS

This retrospective study included adults who visited the ED afebrile at initial presentation. We developed four models based on artificial neural networks to identify suspected serious infection. Patient demographics, vital signs, laboratory test results and information extracted from initial ED physician notes using term frequency-inverse document frequency were used as model variables. Models were trained and internally validated with data from one hospital and externally validated using data from a different hospital. Model discrimination was evaluated using area under the receiver operating characteristic curve (AUC) and 95% confidence intervals (CIs).

RESULTS

The training, internal validation, and external validation datasets comprised 150,699, 37,675, and 85,098 patients, respectively. The AUCs (95% CIs) for Models 1 (demographics + vital signs), 2 (demographics + vital signs + initial ED physician note), 3 (demographics + vital signs + laboratory tests), and 4 (demographics + vital signs + laboratory tests + initial ED physician note) in the internal validation dataset were 0.789 (0.782-0.796), 0.867 (0.862-0.872), 0.881 (0.876-0.887), and 0.911 (0.906-0.915), respectively. In the external validation dataset, the AUCs (95% CIs) of Models 1, 2, 3, and 4 were 0.824 (0.817-0.830), 0.895 (0.890-0.899), 0.879 (0.873-0.884), and 0.913 (0.909-0.917), respectively. Model 1 can be utilized immediately after ED triage, Model 2 can be utilized after the initial physician notes are recorded (median time from ED triage: 28 min), and Models 3 and 4 can be utilized after the initial laboratory tests are reported (median time from ED triage: 68 min).

CONCLUSIONS

We developed and validated models to identify suspected serious infection in the ED. Extracted information from initial ED physician notes using NLP contributed to increased model performance, permitting identification of suspected serious infection at early stages of ED visits.

Modified artificial neural network based on developed snake optimization algorithm for short-term price prediction

Short-term prices prediction is a crucial task for participants in the electricity market, as it enables them to optimize their bidding strategies and mitigate risks. However, the price signal is subject to various factors, including supply, demand, weather conditions, and renewable energy sources, resulting in high volatility and nonlinearity. In this study, a novel approach is introduced that combines Artificial Neural Networks (ANN) with a newly developed Snake Optimization Algorithm (SOA) to forecast short-term price signals in the Nord Pool market. The snake optimization algorithm is utilized to optimize both the structure and weights of the neural network, as well as to select relevant input data based on the similarity of price curves and wind production. To evaluate the effectiveness of the proposed technique, experiments have been conducted using data from two regions of the Nord Pool market, namely DK-1 and SE-1, across different seasons and time horizons. The results demonstrate that the proposed technique surpasses two alternative methods based on Particle Swarm Optimization (PSO) and Genetic Algorithms-based Neural Network (PSOGANN) and Gravitational Search Optimization Algorithm-based Neural Network (GSONN), exhibiting superior accuracy and minimal error rates in short-term price prediction. The results show that the average MAPE index of the proposed technique for the DK-1 region is 3.1292%, which is 32.5% lower than the PSOGA method and 47.1% lower than the GSONN method. For the SE-1 region, the average MAPE index of the proposed technique is 2.7621%, which is 40.4% lower than the PSOGA method and 64.7% lower than the GSONN method. Consequently, the proposed technique holds significant potential as a valuable tool for market participants to enhance their decision-making and planning activities.

Marangoni-driven spreading of a droplet on a miscible thin liquid layer

HYPOTHESIS

The coalescence of droplets with liquid-gas interfaces of different surface tensions is common in nature and industrial applications, where the Marangoni-driven film spreading is an essential process. Unlike immiscible fluids governed by triple contact line dynamics, the mixing between two miscible fluids strongly couples with the film spreading process, which are expected to manifest distinct power-law relations for the temporal increase in the film radius.

EXPERIMENTS

We experimentally investigate the Marangoni-driven film spreading phenomenon for a droplet with lower surface tension dropping onto a miscible, thin liquid layer. The temporal growth of the film radius was detected by using a novel deep convolutional neural network, the U2-net method. Scaling analysis was performed to interpret the spreading dynamics of the film.

FINDINGS

We find that the film radius exhibits a three-stage power-law relation over time, with the exponent varying from 1/2 to 1/8, and back to 1/2. The diffusion-affected Marangoni stresses in these three stages were derived, and two estimations of viscous stress were considered. Through estimating and balancing the viscous stress with the Marangoni stress, the three-stage power-law relation was derived and validated.

Integrating core physics and machine learning for improved parameter prediction in boiling water reactor operations

This study introduces a novel method for enhancing Boiling Water Reactor (BWR) operation simulations by integrating machine learning (ML) models with conventional simulation techniques. The ML model is trained to identify and correct errors in low-fidelity simulation outputs, traditionally derived from core physics computations. These corrections aim to align the low-fidelity results closely with high-fidelity data. Precise predictions of nuclear reactor parameters like core eigenvalue and power distribution are crucial for efficient fuel management and adherence to technical specifications. Current high-fidelity transport calculations, while accurate, are impractical for real-time predictions due to extensive computational demands. Our approach, therefore, utilizes the standard two-step simulation process-assembly-level lattice physics calculations followed by whole-core nodal diffusion computations-to generate initial results, which are then refined using the ML-based error correction model. The methodology focuses on improving simulation accuracy in regular BWR operations rather than developing a universal ML predictor for reactor physics. By training an advanced neural network model on the difference in high-fidelity and low-fidelity simulations, the model can reduce the nodal power error from low-fidelity simulations to around 1% on average and the core eigenvalue down to under 100 pcm. This result is under the condition of the normal variations of control rod pattern and core flow rate changes in standard BWR operations used in the training and evaluation of the machine learning model. This work suggests a promising approach for achieving more accurate, computationally feasible simulation solutions in nuclear reactor operation and management.

Efficient, automatic, and optimized portable Raman-spectrum-based pesticide detection system

Raman spectroscopy can be used for accurately detecting pesticides and determining the chemical composition of a pesticide. To facilitate field detection, the present study used a portable Raman spectrometer for analysis. However, this spectrometer was found to be susceptible to noise interference and signal offsets, which increased the difficulty of pesticide identification. The most commonly used algorithm for Raman spectrum identification is principal component analysis (PCA). However, accurate classification often cannot be achieved with PCA because of the offset and noise in the Raman spectrum data. Therefore, in this study, after the collected Raman spectrum data were processed using the small-step, center-weighted moving-average method, these data were employed to train a convolutional neural network (CNN) model for prediction. To optimize the CNN model, the hyperparameters of the CNN were adjusted using various optimization algorithms, and the optimal solution was obtained after multiple iterations. Data preprocessing and architecture training models were then constructed in a self-optimized manner to improve the ability of the algorithm model to handle diverse types of data. Finally, a CNN model optimized using the cat swarm optimization algorithm was developed. This model was trained on 3000 samples containing three pesticides, and its accuracy for pesticide composition identification was discovered to be 89.33%.

Deep learning solutions for smart city challenges in urban development

In the realm of urban planning, the integration of deep learning technologies has emerged as a transformative force, promising to revolutionize the way cities are designed, managed, and optimized. This research embarks on a multifaceted exploration that combines the power of deep learning with Bayesian regularization techniques to enhance the performance and reliability of neural networks tailored for urban planning applications. Deep learning, characterized by its ability to extract complex patterns from vast urban datasets, has the potential to offer unprecedented insights into urban dynamics, transportation networks, and environmental sustainability. However, the complexity of these models often leads to challenges such as overfitting and limited interpretability. To address these issues, Bayesian regularization methods are employed to imbue neural networks with a principled framework that enhances generalization while quantifying predictive uncertainty. This research unfolds with the practical implementation of Bayesian regularization within neural networks, focusing on applications ranging from traffic prediction, urban infrastructure, data privacy, safety and security. By integrating Bayesian regularization, the aim is to, not only improve model performance in terms of accuracy and reliability but also to provide planners and decision-makers with probabilistic insights into the outcomes of various urban interventions. In tandem with quantitative assessments, graphical analysis is wielded as a crucial tool to visualize the inner workings of deep learning models in the context of urban planning. Through graphical representations, network visualizations, and decision boundary analysis, we uncover how Bayesian regularization influences neural network architecture and enhances interpretability.

An ensemble machine learning approach for the detection of unannounced meals to enhance postprandial glucose control

BACKGROUND

Hybrid automated insulin delivery systems enhance postprandial glucose control in type 1 diabetes, however, meal announcements are burdensome. To overcome this, we propose a machine learning-based automated meal detection approach; METHODS:: A heterogeneous ensemble method combining an artificial neural network, random forest, and logistic regression was employed. Trained and tested on data from two in-silico cohorts comprising 20 and 47 patients. It accounted for various meal sizes (moderate to high) and glucose appearance rates (slow and rapid absorbing). To produce an optimal prediction model, three ensemble configurations were used: logical AND, majority voting, and logical OR. In addition to the in-silico data, the proposed meal detector was also trained and tested using the OhioT1DM dataset. Finally, the meal detector is combined with a bolus insulin compensation scheme; RESULTS:: The ensemble majority voting obtained the best meal detector results for both the in-silico and OhioT1DM cohorts with a sensitivity of 77%, 94%, 61%, precision of 96%, 89%, 72%, F1-score of 85%, 91%, 66%, and with false positives per day values of 0.05, 0.19, 0.17, respectively. Automatic meal detection with insulin compensation has been performed in open-loop insulin therapy using the AND ensemble, chosen for its lower false positive rate. Time-in-range has significantly increased 10.48% and 16.03%, time above range was reduced by 5.16% and 11.85%, with a minimal time below range increase of 0.35% and 2.69% for both in-silico cohorts, respectively, compared to the results without a meal detector; CONCLUSION:: To increase the overall accuracy and robustness of the predictions, this ensemble methodology aims to take advantage of each base model's strengths. All of the results point to the potential application of the proposed meal detector as a separate module for the detection of meals in automated insulin delivery systems to achieve improved glycemic control.

DCDA: CircRNA-Disease Association Prediction with Feed-Forward Neural Network and Deep Autoencoder

Circular RNA is a single-stranded RNA with a closed-loop structure. In recent years, academic research has revealed that circular RNAs play critical roles in biological processes and are related to human diseases. The discovery of potential circRNAs as disease biomarkers and drug targets is crucial since it can help diagnose diseases in the early stages and be used to treat people. However, in conventional experimental methods, conducting experiments to detect associations between circular RNAs and diseases is time-consuming and costly. To overcome this problem, various computational methodologies are proposed to extract essential features for both circular RNAs and diseases and predict the associations. Studies showed that computational methods successfully predicted performance and made it possible to detect possible highly related circular RNAs for diseases. This study proposes a deep learning-based circRNA-disease association predictor methodology called DCDA, which uses multiple data sources to create circRNA and disease features and reveal hidden feature codings of a circular RNA-disease pair with a deep autoencoder, then predict the relation score of the pair by a deep neural network. Fivefold cross-validation results on the benchmark dataset showed that our model outperforms state-of-the-art prediction methods in the literature with the AUC score of 0.9794.

A neural network model for detection and classification of lumbar spinal stenosis on MRI

OBJECTIVES

To develop a three-stage convolutional neural network (CNN) approach to segment anatomical structures, classify the presence of lumbar spinal stenosis (LSS) for all 3 stenosis types: central, lateral recess and foraminal and assess its severity on spine MRI and to demonstrate its efficacy as an accurate and consistent diagnostic tool.

METHODS

The three-stage model was trained on 1635 annotated lumbar spine MRI studies consisting of T2-weighted sagittal and axial planes at each vertebral level. Accuracy of the model was evaluated on an external validation set of 150 MRI studies graded on a scale of absent, mild, moderate or severe by a panel of 7 radiologists. The reference standard for all types was determined by majority voting and in case of disagreement, adjudicated by an external radiologist. The radiologists' diagnoses were then compared to the diagnoses of the model.

RESULTS

The model showed comparable performance to the radiologist average both in terms of the determination of presence/absence of LSS as well as severity classification, for all 3 stenosis types. In the case of central canal stenosis, the sensitivity, specificity and AUROC of the CNN were (0.971, 0.864, 0.963) for binary (presence/absence) classification compared to the radiologist average of (0.786, 0.899, 0.842). For lateral recess stenosis, the sensitivity, specificity and AUROC of the CNN were (0.853, 0.787, 0.907) compared to the radiologist average of (0.713, 0.898, 805). For foraminal stenosis, the sensitivity, specificity and AUROC of the CNN were (0.942, 0.844, 0.950) compared to the radiologist average of (0.879, 0.877, 0.878). Multi-class severity classifications showed similarly comparable statistics.

CONCLUSIONS

The CNN showed comparable performance to radiologist subspecialists for the detection and classification of LSS. The integration of neural network models in the detection of LSS could bring higher accuracy, efficiency, consistency, and post-hoc interpretability in diagnostic practices.

Histology-based quantification of boiling histotripsy outcomes via ResNet-18 network: Towards mechanical dose metrics

This work was focused on the newly developed ultrasonic approach for non-invasive surgery - boiling histotripsy (BH) - recently proposed for mechanical ablation of tissues using pulsed high intensity focused ultrasound (HIFU). The BH lesion is known to depend in size and shape on exposure parameters and mechanical properties, structure and composition of tissue being treated. The aim of this work was to advance the concept of BH dose by investigating quantitative relationships between the parameters of the lesion, pulsing protocols, and targeted tissue properties. A HIFU focus of a 1.5 MHz 256-element array driven by power-enhanced Verasonics system was electronically steered along the grid within 12 × 4 × 12 mm volume to produce volumetric lesions in porcine liver (soft, with abundant collagenous structures) and bovine myocardium (stiff, homogenous cellular) ex vivo tissues with various pulsing protocols (1-10 ms pulses, 1-15 pulses per point). Quantification of the lesion size and completeness was performed through serial histological sectioning, and a computer vision approach using a combination of manual and automated detection of fully fractionated and residual tissue based on neural network ResNet-18 was developed. Histological sample fixation led to underestimation of BH ablation rate compared to the ultrasound-based estimations, and provided similar qualitative feedback as did gross inspection. This suggests that gross observation may be sufficient for qualitatively evaluating the BH treatment completeness. BH efficiency in liver tissue was shown to be insensitive to the changes in pulsing protocol within the tested parameter range, whereas in bovine myocardium the efficiency increased with either increasing pulse length or number of pulses per point or both. The results imply that one universal mechanical dose metric applicable to an arbitrary tissue type is unlikely to be established. The dose metric as a product of the BH pulse duration and the number of pulses per sonication point (BHD1) was shown to be more relevant for initial planning of fractionation of collagenous tissues. The dose metric as a number of pulses per point (BHD2) is more suitable for the treatment planning of softer targets primarily containing cellular tissue, allowing for significant acceleration of treatment using shorter pulses.

Distributed deep reinforcement learning based on bi-objective framework for multi-robot formation

Improving generalization ability in multi-robot formation can reduce repetitive training and calculation. In this paper, we study the multi-robot formation problem with the ability to generalize the target position. Since the generalization ability of neural network is directly proportional to spatial dimension, we adopt the strategy of using different networks to solve different objectives, so that the network learning can focus on the learning of one objective to obtain better performance. In addition, this paper presents a distributed deep reinforcement learning method based on soft actor-critic algorithm for solving multi-robot formation problem. At the same time, the formation evaluation assignment function is designed to adapt to distributed training. Compared with the original algorithm, the improved algorithm can get higher reward cumulative values. The experimental results show that the proposed algorithm can better maintain the desired formation in the moving process, and the rotation design in the reward function makes the multi-robot system have better flexibility in formation. The comparison of control signal curve shows that the proposed algorithm is more stable. At the end of the experiments, the universality of the proposed algorithm in formation maintenance and formation variations is demonstrated.

12-Lead ECG Reconstruction Based on Data From the First Limb Lead

PURPOSE

Electrocardiogram (ECG) data obtained from 12 leads are the most common and informative source for analyzing the cardiovascular system's (CVS) condition in medical practice. However, the large number of electrodes, specific placements on the body, and the need for specialized equipment make the ECG acquisition procedure complex and cumbersome. This raises the challenge of reducing the number of ECG leads by reconstructing missing leads based on available data.

METHODS

Most existing methods for reconstructing missing ECG leads rely on utilizing signals simultaneously from multiple known leads. This study proposes a method for reconstructing ECG data in 12 leads using signal data from the first lead, lead I. Such an approach can significantly simplify the ECG registration procedure. The study demonstrates the effectiveness of using unique models with a developed architecture of artificial neural networks (ANNs) to generate the reconstructed ECG signals. Fragments of ECG from lead I, with a duration of 128 samples and a sampling frequency of 100 Hz, are input to the models. ECG fragments can be extracted from the original signal at arbitrary time points. Each model generates an ECG signal of the same length at its output for the corresponding lead.

RESULTS

Despite existing limitations, the proposed method surpasses known solutions regarding ECG generation quality when using a single lead. The study shows that introducing an additional feature of the direction of the electrical axis of the heart (EAH) as input to the ANN models enhances the generation quality. The quality of ECG generation by the proposed ANN models is found to be dependent on the presence of CVS diseases.

CONCLUSIONS

The developed ECG reconstruction method holds significant potential for use in portable registration devices, screening procedures, and providing support for medical decision-making by healthcare specialists.

A novel method for the natural frequency estimation of the jet engine turbine blades based on its dimensions

This study provides a novel methodology for the natural frequency estimation of the jet engine turbine blade by using the dimension check. This paper presents a summarization of the ongoing research devoted to the method for the turbine blade natural frequency estimation. The main target of the research presented in the paper is to develop a novel method that can calculate the natural frequency of a particular turbine blade by using the dimensions of investigated turbine blade from a dimension check. This goal is achieved by the combination and interaction of several methods as for instance computed aided design (CAD) finite element modelling (FEM), artificial neural network (ANN) and others. As it is mentioned in the following chapters of the article a unique novel method is developed that can predict natural frequency according to the dimensions. The results confirmed the correctness of the new methodology, which can predict natural frequency by the dimensions of a turbine blade immediately with a relatively high level of accuracy (maximal errors are under 1.5%). Every jet engine manufacturer (GE aviation, Rolls Royce, Prat and Whitney, etc.) has to test jet engine parts for the natural frequencies in order to avoid the resonance at early stage of the manufacturing process in order to mount the blades into the engine. The experimental tests of every single turbine blade are time-consuming, a novel method can predict natural frequency according to the dimensions by using data from dimension check in 0.0051 s. The presented method is under patent pending.

Automated segmentation of lesions and organs at risk on [68Ga]Ga-PSMA-11 PET/CT images using self-supervised learning with Swin UNETR

BACKGROUND

Prostate-specific membrane antigen (PSMA) PET/CT imaging is widely used for quantitative image analysis, especially in radioligand therapy (RLT) for metastatic castration-resistant prostate cancer (mCRPC). Unknown features influencing PSMA biodistribution can be explored by analyzing segmented organs at risk (OAR) and lesions. Manual segmentation is time-consuming and labor-intensive, so automated segmentation methods are desirable. Training deep-learning segmentation models is challenging due to the scarcity of high-quality annotated images. Addressing this, we developed shifted windows UNEt TRansformers (Swin UNETR) for fully automated segmentation. Within a self-supervised framework, the model's encoder was pre-trained on unlabeled data. The entire model was fine-tuned, including its decoder, using labeled data.

METHODS

In this work, 752 whole-body [68Ga]Ga-PSMA-11 PET/CT images were collected from two centers. For self-supervised model pre-training, 652 unlabeled images were employed. The remaining 100 images were manually labeled for supervised training. In the supervised training phase, 5-fold cross-validation was used with 64 images for model training and 16 for validation, from one center. For testing, 20 hold-out images, evenly distributed between two centers, were used. Image segmentation and quantification metrics were evaluated on the test set compared to the ground-truth segmentation conducted by a nuclear medicine physician.

RESULTS

The model generates high-quality OARs and lesion segmentation in lesion-positive cases, including mCRPC. The results show that self-supervised pre-training significantly improved the average dice similarity coefficient (DSC) for all classes by about 3%. Compared to nnU-Net, a well-established model in medical image segmentation, our approach outperformed with a 5% higher DSC. This improvement was attributed to our model's combined use of self-supervised pre-training and supervised fine-tuning, specifically when applied to PET/CT input. Our best model had the lowest DSC for lesions at 0.68 and the highest for liver at 0.95.

CONCLUSIONS

We developed a state-of-the-art neural network using self-supervised pre-training on whole-body [68Ga]Ga-PSMA-11 PET/CT images, followed by fine-tuning on a limited set of annotated images. The model generates high-quality OARs and lesion segmentation for PSMA image analysis. The generalizable model holds potential for various clinical applications, including enhanced RLT and patient-specific internal dosimetry.

Modeling of changes in the nuclide composition of VVER reactor fuel using artificial neural network

The paper seeks to give a computer-based model, using an artificially intelligent technique. This is the adaptive neuro-fuzzy inference system (ANFIS), to predict the concentration changes of certain nuclides in the uranium fuel used for VVER reactors. It uses low-enriched uranium dioxide as a fuel in its solid state. The reactivity in the core is controlled by the control rods. Nuclide concentration changes in the reactor fuel, if not monitored, may cause the unsafe operation of the reactor. Hence, the need for this study. The nuclides considered in this study are, U-235, U-236, U-237, U-238, Pu-239, Pu-240, Pu-241, Am-242 and Am-243. The initial computational technique was performed using MATLAB Simulink. The simulation data for all the concentrations of the nuclides were obtained. Then the proposed ANFIS model was performed and tested using data from the Simulink. Results from the simulink and ANFIS were compared and the results were in good agreement. Again, the results were compared to the Calculating Actinide Inventory (CAIN) code from the IAEA-TECHDOC-1535 published in 2007 and both showed a good agreement. An RMSE of about 0.98% and 1.25% were obtained for training and testing data respectively. The developed model will allow technologists to quickly perform calculations for the reactor, which is essential for safety systems. It could be concluded that the ANFIS model can effectively be used to predict the concentration of each nuclide in the uranium fuel because it is effective, precise with lesser error, and does not consume time.

LFC-UNet: learned lossless medical image fast compression with U-Net

In the field of medicine, the rapid advancement of medical technology has significantly increased the speed of medical image generation, compelling us to seek efficient methods for image compression. Neural networks, owing to their outstanding image estimation capabilities, have provided new avenues for lossless compression. In recent years, learning-based lossless image compression methods, combining neural network predictions with residuals, have achieved performance comparable to traditional non-learning algorithms. However, existing methods have not taken into account that residuals often concentrate excessively, hindering the neural network's ability to learn accurate residual probability estimation. To address this issue, this study employs a weighted cross-entropy method to handle the imbalance in residual categories. In terms of network architecture, we introduce skip connections from U-Net to better capture image features, thereby obtaining accurate probability estimates. Furthermore, our framework boasts excellent encoding speed, as the model is able to acquire all residuals and residual probabilities in a single inference pass. The experimental results demonstrate that the proposed method achieves state-of-the-art performance on medical datasets while also offering the fastest processing speed. As illustrated by an instance using head CT data, our approach achieves a compression efficiency of 2.30 bits per pixel, with a processing time of only 0.320 seconds per image.

Time-varying gain extended state observer-based adaptive optimal control for disturbed unmanned helicopter

In this paper, the robust adaptive optimal tracking control problem is addressed for the disturbed unmanned helicopter based on the time-varying gain extended state observer (TVGESO) and adaptive dynamic programming (ADP) methods. Firstly, a novel TVGESO is developed to tackle the unknown disturbance, which can overcome the drawback of initial peaking phenomenon in the traditional linear ESO method. Meanwhile, compared with the nonlinear ESO, the proposed TVGESO possesses easier and rigorous stability analysis process. Subsequently, the optimal tracking control issue for the original unmanned helicopter system is transformed into an optimization stabilization problem. By means of the ADP and neural network techniques, the feedforward controller and optimal feedback controller are skillfully designed. Compared with the conventional backstepping approach, the designed anti-disturbance optimal controller can make the unmanned helicopter accomplish the tracking task with less energy. Finally, simulation comparisons demonstrate the validity of the developed control scheme.

Deep neural networks for crack detection inside structures

Crack detection is a long-standing topic in structural health monitoring. Conventional damage detection techniques rely on intensive, time-consuming, resource-intensive intervention. The current trend of crack detection emphasizes using deep neural networks to build an automated pipeline from measured signals to damaged areas. This work focuses on the seismic-wave-based technique of crack detection for plate structures. Previous work proposed an encoder-decoder network to extract crack-related wave patterns from measured wave signals and predict crack existence on the plate. We extend previous work with extensive experiments on different network components and a data preprocessing strategy. The proposed methods are tested on an expanded crack detection dataset. We found that a robust backbone network, such as Densely Connected Convolutional Network (DenseNet) can effectively extract the features characterizing cracks of wave signals, and by using the reference wave field for normalization, the accuracy of detecting small cracks can be further improved.

Ensemble averaging deep neural network for botnet detection in heterogeneous Internet of Things devices

The botnet attack is one of the coordinated attack types that can infect Internet of Things (IoT) devices and cause them to malfunction. Botnets can steal sensitive information from IoT devices and control them to launch another attack, such as a Distributed Denial-of-Service (DDoS) attack or email spam. This attack is commonly detected using a network-based Intrusion Detection System (NIDS) that monitors the network device's activity. However, IoT network is dynamic and IoT devices have many types with different configurations and vendors in IoT environments. Therefore, this research proposes an Intrusion Detection System (IDS) by ensemble-ing traffic from heterogeneous IoT devices. This research proposes Deep Neural Network (DNN) to create a training model from each heterogeneous IoT device. After that, each training model from each heterogeneous IoT device is used to predict the traffic. The prediction results from each training model are averaged using the ensemble averaging method to determine the final result. This research used the N-BaIoT dataset to validate the proposed IDS model. Based on experimental results, ensemble averaging DNN can detect botnet attacks in heterogeneous IoT devices with an average accuracy of 97.21, precision of 91.41, recall of 87.31, and F1-score 88.48.

Construction and optimization of vending machine decision support system based on improved C4.5 decision tree

The intensification of market competition makes refined operation management become the focus of attention of major manufacturers. As an important branch of artificial intelligence (AI), machine learning (ML) plays a key role in it, and has its application prospect in various systems. Based on this situation, this paper takes vending machines as the research object. On the one hand, the product classification model of vending machine is constructed based on decision tree algorithm. On the other hand, based on neural network (NN), the sales forecast model of vending machines is built. Finally, based on the above research, the theoretical framework of decision support system (DSS) for vending machines is constructed. The research shows that: (1) The accuracy of C4.5 algorithm can reach 87 % at the highest and 68 % at the lowest. The accuracy of the improved C4.5 algorithm can reach 87 % at the highest and 67 % at the lowest, with little difference between them. (2) The maximum running time of the improved C4.5 algorithm is about 5500 ms, and the minimum is close to 1 ms. In addition, the running time of all seven datasets is better than that of the unmodified algorithm. (3) When the back propagation neural network (BPNN) is used to forecast the sales of vending machines, the curve of the predicted data basically coincides with the curve of the actual data, which shows that its accuracy is high. This paper aims to build a convenient and secure DSS by taking vending machines as an example. In addition, this paper also uses reinforcement learning to optimize the research methods of this paper. It can further optimize the performance and efficiency of vending machines and provide better service experience for customers. Meanwhile, the use of reinforcement learning can make the whole system more intelligent and adaptive to better cope with the changing market environment.