Fusion of transfer learning models for detection of alzheimer's disease using bidirectional long short-term memory with equilibrium optimization algorithm.

Nucleic acids, the fundamental building blocks of life, serve as versatile tools in genetic information retrieval, disease diagnosis, and biotechnological applications. The automated Intelligent Robotic System for Nucleic Acid Extraction and Library Preparation (iRoNAEaLP) tool represents a significant advancement in nucleic acid extraction and library preparation in an automated manner, addressing complexity and diversity while minimizing human involvement. Utilising machine learning algorithms and a Long Short-Term Memory (LSTM) architecture, iRoNAEaLP autonomously generates process flowcharts, predetermined reagents and consumable quantities, and aligns process steps with specific module actions via strategy-guided segmented program file arrangements. As a result, the biological outcome from this system has demonstrated high efficiency and large-scale data quality in various types of samples in terms of trace nucleic acid extraction, plasmid/genetic construct extraction, and single-cell and spatial omics, which require mRNA library preparation for smart-seq2 sequencing. This innovation paves the way for more efficient and accessible bioprocesses in various life science applications.

In the modern world, people worldwide face different forms of depression due to factors such as workplace stress, economic pressures, and other causes. The rise of Artificial Intelligence (AI) has enabled data analysis and solving of real-world problems. People frequently use social media platforms to communicate and express their feelings. Hence, social media data is helpful for research purposes, particularly for automatic depression detection. Numerous scholarly works have explored using learning-based approaches to identify sadness from social media interactions. However, individual existing deep learning models have limitations, such as the inability to capture contextual and sequential dependencies in text fully. We addressed this by proposing a deep learning-based, non-invasive approach to identify depression in social media conversations. Our proposed approach involves a novel hybrid deep learning model, Depression Detection Network (DDNet), which combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models. The model was trained and tested on a manually annotated dataset of 8500 depression-related tweets (6,800 for training and 1,700 for testing) collected via the Twitter Application Programming Interface (API). The DDNet model achieved a high accuracy of 96.21%, outperforming baseline models such as standalone LSTM (92.31%) and Recurrent Neural Network (RNN) (91.43%). Furthermore, we developed Hybrid Deep Learning-based Depression Detection (HDL-DD), an algorithm that processes social media text and predicts potential depressive tendencies. The experimental results indicate that DDNet significantly improves depression classification, achieving 95% precision, 96% recall, and 95% F1-score, demonstrating its effectiveness over existing methods. By recognizing depression with a 96.21% accuracy rate, our deep learning model outperformed previous state-of- the-art approaches, making it a promising tool for automated depression monitoring applications. This approach could be integrated into real-world social media-based mental health monitoring applications, supporting early intervention efforts and contributing to AI-driven healthcare solutions.

AI-driven transcriptomic biomarker discovery for early identification of pediatric deterioration in Acute Care.

Pre-trained large language models outperform statistics and machine learning forecasting visits in the emergency departments.

The natural gas supply crisis triggered by the Russia–Ukraine conflict has laid bare the energy market’s extreme vulnerability in the face of geopolitical risk, highlighting the need for accurate multi-step gas price forecasting. However, most AI-based energy price studies have a gap: they focus on single-step prediction or homogeneous model comparisons, lacking analysis of performance degradation in multi-step dynamic frameworks. This study takes daily natural gas price data from the Henry Hub in the United States from 1997 to 2024 as the research object, constructs a multi-step prediction framework with a step size ranging from 1 to 4 days, and systematically compares the prediction performances of four artificial intelligence models: feedforward neural network, support vector machine, random forest, and long short-term memory network. The quantitative results show that, across all prediction cycles, the long short-term memory model has the lowest error rate. For example, in one-step forecasting, its Mean Absolute Percentage Error is 8.53%. Practically, the findings matter. Taking European governments facing natural gas shortages in the Russia-Ukraine conflict as an example, LSTM models can be used for multi-step prediction to forecast price fluctuations 2–4 days in advance, optimizing import reserve strategies to avoid supply disruptions; energy traders can use this to design robust futures arbitrage portfolios. In summary, the research provides a scientific basis and reference for government energy security policy-making and institutional investor trading.

To automate the classification of functional brain networks in epilepsy patients using resting-state functional magnetic resonance imaging (rs-fMRI). The study introduces a deep learning framework that leverages spatial and temporal features to classify Independent Component Analysis (ICA)-derived networks into 11 functionally distinct classes, including seizure onset zone (SoZ), resting-state networks (RSNs), and artifact/noise. A hybrid deep learning architecture was developed combining a 3D Convolutional Neural Network (3D-CNN) to extract spatial features (SF) and a Long Short-Term Memory (LSTM) network to capture temporal dynamics from time-domain (TS) and frequency-domain (FS) signals. These multi-domain features were concatenated and classified into 11 distinct ICA component types. An ablation study assessed the individual and combined contributions of spatial, temporal, and spectral features. Additionally, expert neurologists independently rated four representative cases to qualitatively validate the model's interpretability and clinical relevance. The baseline 3D CNN (SF) model achieved an overall accuracy of 69% with a sensitivity of 0.52 and a ROC AUC of 0.76. Incorporating frequency-domain signals (SF + FS) enhanced sensitivity to 0.54 and improved the ROC AUC to 0.78 while maintaining a similar accuracy. Combining both time-domain and frequency-domain signals (SF + TS + FS) yielded the highest accuracy at 70%. At the class level, the Noise class consistently demonstrated robust performance (up to 0.94), whereas the temporal lobe network class Temporal class exhibited lower scores (0.14-0.24) across all configurations. Our results demonstrate that this data-driven framework can effectively automate the classification of rs-fMRI-derived functional brain networks including SoZ thereby reducing subjectivity and workload in clinical review. The inclusion of spatial, temporal, and spectral information enables a richer and more nuanced classification that supports downstream applications in epilepsy surgical planning.

Bidirectional long short-term memory with style-based recalibration module for automated bundle branch block detection

Food price stabilization remains a critical challenge in economic development planning and food security, particularly in developing countries like Indonesia, which exhibit high spatial and temporal diversity. To develop an efficient and adaptive predictive approach for understanding food commodity price dynamics, this study integrates multivariate time series clustering using a Dynamic Time Warping-based K-Means algorithm with a hybrid forecasting model that combines Vector Error Correction Model with Exogenous Variables and Long Short-Term Memory. The clustering evaluation results indicate reasonably cohesive group structures, with a silhouette score of 0.45 and a Davies-Bouldin Index of 0.67. Each cluster profile reveals significant differences in price trends, volatility, and anomaly patterns. Model validation using the Wilcoxon signed-rank test shows that the differences between cluster-level forecasts and individual-level actual values are generally statistically insignificant. These findings suggest that the proposed integrative approach can accurately capture regional price patterns and serve as a foundation for more data-driven and responsive policymaking in food price stabilization efforts. The 30-period forecasts for rice, eggs, and red onions reflected dynamic variations aligned with spatial characteristics: rice shows relatively stable behavior, eggs exhibit strong seasonal patterns, and red onions display the highest price volatility.

Does brain connectivity hold the key to safer roads? EEG-based fatigue detection in young drivers using interpretable deep learning.

EEG signal is being widely used in the field of emotion recognition, which currently suffers from the difficulty of obtaining highly distinguishable features. We propose CNN-BiLSTM-CS for emotion recognition EEG-based, which is to address the shortcomings of the traditional LSTM unidirectional propagation and Softmax supervised model in feature extraction. The method firstly employs BiLSTM to CNN, which can bilaterally obtain emotion feature information, and then introduces Center and Softmax to form a joint loss function to minimize the intra-class distance and maximize the inter-class distance, which can improve the recognition ability. DEAP and SEED dataset are employed to test the performance of CNN-BiLSTM-CS. The results of the average accuracy of valence and arousal are 94.22% and 92.16% on DEAP, which is increase by almost 6% to CNN-LSTM. The triple categorization accuracy of the SEED dataset is 95.45%. CNN-BiLSTM-CS significantly improves the recognition performance of deep features of EEG through the improved network structure and combined loss function.

Harnessing artificial intelligence for dengue forecasting in climate-vulnerable regions: A narrative review with insights from Ba Ria-Vung Tau, Vietnam.

RimeSleepNet: A hybrid deep learning network for s-EEG sleep stage classification.

Recent research highlights the potential of ultrasound (US) stimulation as a noninvasive tool for modulating neural and cellular signaling in the spleen and liver to treat inflammatory diseases and diabetes. However, challenges like nerve activation failures, off-target stimulation, and organ motion during respiration can affect treatment efficacy. This study introduces a novel tracking framework for accurate liver and spleen motion tracking using US imaging to overcome these challenges. The tracking framework integrates an enhanced Kanade-Lucas-Tomasi (EKLT) tracker with a long short-term memory (LSTM) predictor. The EKLT tracker provides precise annotations that improve LSTM training, while the LSTM compensates for occlusions and noise by making predictions based on prior data and dynamically adjusting the region of interest (ROI). Spleen motion tracking was evaluated using 40 recordings from 10 participants, each undergoing four distinct breathing patterns. Additionally, the method was evaluated on a liver motion dataset from MICCAI, collected from 9 subjects. Spleen tracking was most accurate during slow, shallow breathing, with an average error of 0.4 $\pm$ 0.4 mm, and had an average error of 1.37 $\pm$ 0.9 mm during fast, deep breathing. Liver tracking results showed high accuracy with an average error of 0.3 $\pm$ 0.2 mm. The EKLT-LSTM framework offers advantages over previous tracking models, providing high accuracy in tracking liver and spleen motion under occlusion and noisy conditions. The EKLT-LSTM is suitable for end-organ modulation applications and can be adapted to other ultrasound-guided therapies and bioelectronic medicine.

A novel MRI image-analyzed stroke prediction framework using multiscale residual long short-term memory with spatial attention network

A significant obstacle to preserving grid stability and incorporating renewable energy into smart grids is variations in solar irradiation. To improve solar power management's dependability, this research proposes a short-term solar forecasting framework powered by AI. Multiple machine learning models, such as long short-term memory (LSTM), random forest (RF), gradient boosting (GB), AdaBoost, neural networks (NN), K-Nearest neighbor (KNN), and linear regression (LR), are integrated into the suggested system, which also uses principal component analysis (PCA) for dimensionality reduction. The Abiod Sid Cheikh station in Algeria (2019-2021) provided real-world data for the model's validation. With a two-hour-ahead RMSE of 0.557 kW/m², AdaBoost had the most accuracy, whereas LR had the lowest, at 0.510 kW/m². In addition to increasing computing efficiency, PCA preserved 99.3% of the data volatility. In addition to increasing computing efficiency, PCA preserved 99.3% of the data volatility. These findings highlight the efficiency of hybrid AI models based on PCA for accurate forecasting, which is crucial for smart grid stability.

HCNS:A deep learning model for identifying essential proteins based on hypergraph convolution and sequence features.

<p>Progress in mobile technology, the internet, cloud computing, digital platforms, and social media has substantially facilitated interpersonal connections following the COVID-19 pandemic. As individuals increasingly prioritise health, there is an escalating desire for novel methods to assess health and well-being. This study presents an internet of things (IoT)-based system for remote monitoring utilizing a long range (LoRa), a low-cost and LoRa wireless network for the early identification of health issues in home healthcare environments. The project has three primary components: transmitter, receiver, and alarm systems. The transmission segment captures data via sensors and transmits it to the reception segment, which then uploads it to the cloud. Additionally, machine learning (ML) methods, including convolutional neural networks (CNN), artificial neural networks (ANN), Naïve Bayes (NB), and long short-term memory (LSTM), were utilized on the acquired data to forecast heart rate, blood oxygen levels, body temperature patterns. The forecasting models are trained and evaluated using data from various health parameters from five diverse persons to ascertain the architecture that exhibits optimal performance in modeling and predicting dynamics of different medical parameters. The models' accuracy was assessed using mean absolute error (MAE) and root mean square error (RMSE) measures. Although the models performed similarly, the ANN model outperformed them in all conditions.</p>

Fertilizer prediction using serial exponential newton meta-heuristic algorithm-based convolutional neural network in IoT-based WSNs.

Fabric tearing performance testing experiment is an important part of evaluating fabric durability. The aim of this paper is to solve the problem of real-time prediction of fabric tearing performance testing by effectively extracting key features from experimental data and constructing a prediction model applicable to the process of fabric tearing performance testing. In this study, the trend prediction model for the experimental process of fabric tear performance testing (BLTT-FT) based on the “bidirectional long- and short-term attention mechanism” is adopted. A prediction model combining the improved Bi-directional Long Short-Term Memory (BiLSTM) structure, Transformer encoding layer, and Temporal Convolutional Network (TCN) layer is proposed. While considering sequence information globally, the model captures the bidirectional dependence of time series, reduces model complexity through the TCN layer, and finally optimizes prediction accuracy via the fully connected layer and activation function, thus achieving multi-step prediction. Analysis of variance (ANOVA) indicates that, across multiple datasets constructed from fabrics with different elasticity grades, the model shows extremely significant differences (p < 0.001) in the metrics of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) at each prediction step. Furthermore, it maintains a low error level even in the long-range prediction scope: the average RMSE of multi-step prediction is 0.0881, the average MAE of multi-step prediction is 0.0609, the average MAPE of multi-step prediction is as low as 3.06%, and the average coefficient of determination (R2) of multi-step prediction is as high as 0.9572. The ablation experiments confirm that multi-modular hierarchical modeling effectively solves the problem of detail accuracy of single-step prediction and long-range dependence of multi-step prediction. The results show that the proposed model performs well in real-time trend prediction results for different data sets constructed from fabrics with different elasticity grades. By predicting the dynamics of the experimental process of fabric tearing performance testing in real time, this study has exploratory value in improving the experimental efficiency and optimizing the experimental process.