Search Results (707)

No systematic review has previously examined the application of large language models (LLMs) for generating impressions from radiology report findings. This study systematically reviews the performance of LLMs on this task and their associated evaluation methodologies. A search of seven electronic databases on 7 August 2025 identified 15 eligible papers (average quality score: 71.4%). These articles evaluated 35 LLMs, including 21 base models. The reported performance ranges were as follows: Recall-Oriented Understudy for Gisting Evaluation (ROUGE)-1, 35.9% (Generative Pre-Trained Transformer (GPT)-4) to 69.7% (Baichuan2-13B); ROUGE-2, 13.4% (Large Language Model Meta AI (Llama)) to 52.4% (Baichuan2-13B); and ROUGE-L, 16.5% (Chat General Language Model–Medical (ChatGLM-Med)) to 63.8% (finetuned Text-to-Text Transfer Transformer (T5)). The finetuned T5 consistently demonstrated high performance, based on Bidirectional Encoder Representations from Transformers Score (BERTScore): 89.2%; BiLingual Evaluation Understudy (BLEU)-1: 65.2%; BLEU-2: 57.9%; BLEU-3: 52.5%; BLEU-4: 48.3%; Metric for Evaluation of Translation with Explicit ORdering (METEOR): 38.1%; ROUGE-1: 59.9%; ROUGE-2: 50.9%; ROUGE-L: 63.8%; and subjective metrics (clinical usability: 4.5/5.0; completeness: 4.3/5.0; conciseness: 4.3/5.0; fluency: 4.4/5.0). These results, based on 132,043 computed tomography, echocardiography, magnetic resonance imaging, and X-ray reports, indicate its strong clinical potential for assisting radiologists in impression generation through supervised finetuning rather than prompting techniques used in closed-source LLMs. Full article

Whole-body vibration (WBV) remains a critical factor influencing ride comfort, driver performance and occupational health in vehicle applications. Despite the widespread use of standardized indicators, assessing WBV exposure and its perceptual implications remains challenging due to the complex interaction between road excitation, vehicle dynamics, seat transmissibility and human biodynamic response. This review provides a comprehensive synthesis of contemporary methods for WBV assessment, emphasizing their theoretical foundations, practical implementation and inherent limitations. The paper examines classical evaluation metrics, including frequency-weighted root mean square acceleration and vibration dose value, alongside complementary approaches such as overall vibration total value, absorbed power and motion sickness indicators. Biodynamic modeling strategies for the human–seat–vehicle system are critically reviewed, highlighting trade-offs between model simplicity and physiological realism. Particular attention is given to road surface representation and excitation modeling, discussing the implications of ISO 8608-based stochastic profiles versus measured, time-domain inputs on WBV assessment outcomes. Simulation frameworks, experimental platforms and driving simulators are reviewed as complementary tools for evaluating vibration exposure and validating predictive models. Emerging methods, including time–frequency analysis and data-driven approaches, are discussed with a focus on interpretability, validation and integration with established standards such as ISO 2631. The review consolidates recent advances in integrated evaluation approaches, including the role of driving simulators and simulation-, hardware- and driver-in-the-loop (SiL/HiL/DiL) frameworks as enabling tools for repeatable testing, objective–subjective comfort correlation and early-stage vibration-control development. By critically examining both established and emerging methodologies, this review aims to support informed selection and interpretation of WBV assessment tools in vehicle design and evaluation. The findings underscore the need for integrated, transparent and application-oriented approaches to advance vibration comfort assessment and guide future research and standardization efforts. Full article

Accurate solar forecasting is critical for photovoltaic integration, particularly in regions with complex terrain and limited observations. This study benchmarks five deep learning architectures—MLP, RNN, LSTM, CNN, and a Grey Wolf Optimizer–enhanced MLP (MLP–GWO)—to evaluate the accuracy–computational efficiency trade-off for generating daily solar potential maps from ERA5 reanalysis over Mexico. Models were trained using a strict temporal split on a high-dimensional grid (85 × 129 points, flattened to 10,965 outputs) and evaluated in terms of predictive skill and hardware cost. The RNN achieved the best overall performance (RMSE ≈ 32.3, MAE ≈ 27.8, R² ≈ 0.96), while the MLP provided a competitive lower-complexity alternative (RMSE ≈ 54.8, MAE ≈ 46.0, R² ≈ 0.88). In contrast, the LSTM and CNN showed poorer generalization, and the MLP–GWO failed to converge. For the CNN, this underperformance is linked to the intentionally flattened spatial representation. Overall, the results indicate that within a specific ERA5-based, daily-resolution, and resource-constrained experimental framework, lightweight architectures such as RNNs and MLPs offer the most favorable balance between accuracy and computational efficiency. These findings position them as efficient surrogates of ERA5-derived daily solar potential suitable for large-scale, preliminary energy planning applications. Full article

The use of virtual reality for pilot flight training, whether as a stand-alone device, or to augment or replace a conventional simulator, has gained significant attention in recent years. The primary purported benefit of virtual reality is its increased ability to achieve immersion of the trainee, which has particular benefits for visuospatial awareness. This benefit of the technology would appear to offer little advantage in the training of instrument-flying skills, where only the aircraft’s instrumentation needs to be accurately rendered in order that the status of the ownship can be known. However, given the wide-scale intention toward the adoption of the technology, it is likely that instrument flight training will be one of its uses at flight schools. In order that the effectiveness of the VR Simulator can be evaluated, for instrument flight training, a quasi-randomised separate-sample pretest–posttest design study was completed. The ability of this low-cost VR simulator to transfer the flying skills required to conduct an ILS approach, after establishment on approach, was evaluated with 44 participants. Results indicate significant improvement in participants’ flying skills based on operational (r_rb = 0.508) and synthetic (g = 0.844) performance metrics. The findings indicate that the VR simulator appears effective for the training of these skills, and that the immersion and presence are not detrimental, even when the primary focus is the instrument panel. The idea that VR is an effective tool for training instrument flight skills has not previously been demonstrated. Due consideration must, however, be given to the context of this study and the noted limitations of the VR technology. Full article

The increasing penetration of distributed generation in active distribution networks (ADNs) introduces significant voltage regulation challenges due to the intermittent nature of renewable energy sources. Electric springs (ESs) have emerged as a cost-effective alternative to conventional FACTS devices for voltage regulation, requiring minimal energy storage while providing fast, flexible reactive power compensation. This paper proposes a deep reinforcement learning (DRL)-based approach for voltage regulation in balanced active distribution networks with distributed generation. Electric springs are deployed at selected buses in series with noncritical loads to provide flexible voltage support. The main contributions of this work are: (1) a novel region-based penalized reward function that effectively guides the DRL agent to minimize voltage deviations; (2) a coordinated control strategy for multiple ESs using the Deep Deterministic Policy Gradient (DDPG) algorithm, representing the first application of DRL to ES-based voltage regulation; (3) a systematic hyperparameter tuning methodology that significantly improves controller performance; and (4) comprehensive validation demonstrating an approximately 40% reduction in mean voltage deviation relative to the no-control baseline. Three well-known continuous-control DRL algorithms, Twin Delayed Deep Deterministic Policy Gradient (TD3), Proximal Policy Optimization (PPO), and DDPG, are first evaluated using the default hyperparameter configurations provided by MATLAB R2022b.Based on this baseline comparison, a dedicated hyperparameter-tuning procedure is then applied to DDPG to improve the robustness and performance of the resulting controller. The proposed approach is evaluated through simulation studies on the IEEE 33-bus and IEEE 69-bus test systems with time-varying load profiles and fluctuating renewable generation scenarios. Full article

Graph anomaly detection (GAD) is pivotal for security-critical applications like cybersecurity and financial fraud detection. While reconstruction-based Graph Neural Networks (GNNs) are prevalent, their efficacy is often compromised by two phenomena: (1) anomaly overfitting, where expressive models capture anomalous patterns, and (2) homophily-induced attenuation, where message passing smooths localized anomaly cues. This paper proposes CLEAN-GAD, a contamination-aware framework that mitigates anomaly influence during training through multi-view robust learning. Specifically, we develop a contrastive augmentation module that utilizes local inconsistency scores to identify and suppress pseudo-anomalous nodes and edges, thereby yielding a purified augmented view. To capture diverse anomaly signals, a frequency-adaptive encoder with dual-pass channels is designed to integrate low- and high-frequency information. Furthermore, we introduce a distribution-separation regularizer and cross-view alignment to stabilize learning and resolve view shifts. Theoretical analysis confirms that reducing the contamination ratio $\rho$ expands the reconstruction-risk gap between normal and anomalous nodes, inherently boosting detection performance. Extensive experiments on multiple benchmark datasets from various domains demonstrate the superior anomaly detection performance of CLEAN-GAD. Full article

High-dimensional tabular data are common in biomedical and clinical research, yet conventional machine learning methods often struggle in such settings due to data scarcity, feature redundancy, and limited generalization. In this study, we systematically evaluate Synolitic Graph Neural Networks (SGNNs), a framework that transforms high-dimensional samples into sample-specific graphs by training ensembles of low-dimensional pairwise classifiers and analyzing the resulting graph structure with Graph Neural Networks. We benchmark convolution-based (GCN) and attention-based (GATv2) models across 15 UCI datasets under two training regimes: a foundation setting that concatenates all datasets and a dataset-specific setting with macro-averaged evaluation. We further assess cross-dataset transfer, robustness to limited training data, feature redundancy, and computational efficiency, and extend the analysis to a real-world ovarian cancer proteomics dataset. The results show that topology-aware node feature augmentation provides the dominant performance gains across all regimes. In the foundation setting, GATv2 achieves an ROC-AUC of up to 92.22 (GCN: 91.22), substantially outperforming XGBoost (86.05), $\alpha =0.001$. In the dataset-specific regime, GATv2, combined with minimum-connectivity filtering, achieves a macro ROC-AUC of 83.12, compared to 80.28 for XGBoost. Leave-one-dataset-out evaluation confirms cross-domain transfer, with an ROC-AUC of up to 81.99. SGNNs maintain ROC-AUC around 85% with as little as 10% of the training data and consistently outperform XGBoost in more extreme low-data regimes, $\alpha =0.001$. On ovarian cancer proteomics data, foundation training improves both predictive performance and stability. Efficiency analysis shows that graph filtering substantially reduces training time, inference latency, and memory usage without compromising accuracy. Overall, these findings suggest that SGNNs provide a robust and scalable approach for learning from high-dimensional, heterogeneous tabular data, particularly in biomedical settings with limited sample sizes. Full article

The growing miniaturization of electronic systems and the expansion of sustainable, autonomous IoT technologies emphasize the need for efficient, ultra-low-power energy harvesting devices. This study evaluates fifteen devices from five industry leaders for use in small-scale autonomous seed germination systems. Its novelty lies in applying a competitive profile matrix within a flexible multicriteria evaluation framework based on the simple additive weighting (SAW) method that uses a comprehensive set of competitive technology factors (CTFs). The results demonstrate that a transparent and structured methodology can generate prioritized lists of suitable energy harvesters while accounting for technical, economic, and environmental trade-offs. The study also shows that device rankings depend on the scope and objectives of the project. If these change, then the CTF selection, classification, and weighting adjust accordingly. Therefore, the relevance of this study lies in the adaptability, replicability, and audibility of the proposed framework, which supports the selection of informed technology for autonomous, IoT-based germination systems and other technological projects. Full article

Mechatronics developed under the influence of the Third Industrial Revolution and was a discipline that provided methods and tools for the development of industrial robots, advanced machine tools, mobile phones, and automobiles, among other sophisticated products. With the emergence of Industry 4.0 in 2011, mechatronics has become indispensable, as traditional production systems are being transformed into cyber-physical systems (CPS), some of which are composed of sophisticated technologies such as Digital Twins (DT) and sophisticated robots, among others. In 2020, the Fifth Industrial Revolution began, giving rise to so-called Human Cyber-Physical Systems (HCPS) and promoting the use of Cobots in industries. Because today’s industrial world is influenced by three active industrial revolutions and two transitions, it is possible to find machines and production systems that were designed with different principles and for different purposes, making it necessary to propose a classification that allows each system to be located according to the premises of its respective industrial revolution. This article analyzes the evolution of mechatronics and proposes a classification of machines and production systems based on the premises of each industrial revolution. The objective is to determine the influence of mechatronics on the different types of machines that exist today and analyze its implications. Full article

Technologies are the enablers of the practical realization of benefits from fundamentals originating from various disciplines of science and engineering for a holistic improvement in the life of livings (plants, animal and humans all included), development of rational societies and the health of planet Earth [...] Full article

To address the limitations of traditional aircraft skin inspection, the aviation industry and academia have increasingly been exploring the integration of computer vision technologies into the defect detection process. These implementations of computer vision technologies rely on the performance of underlying neural network models, whose effectiveness is highly influenced by their hyperparameter configuration. To obtain optimum hyperparameters, an optimization procedure is often employed to optimize a certain combination of the model’s performance metrics. However, in the aircraft skin defect detection domain, studies to inspect the effect of different emphases in the performance metrics considered in this objective function are still not widely available. In this paper, we present our early observations regarding the influence of different performance metrics’ emphases during the hyperparameter tuning process on the overall performance of a computer vision model employed for aircraft skin defect detection. In this preliminary inspection, we consider the utilization of YOLOv12 and the Bayesian Optimization approach for the defect detection model and hyperparameter optimizer, respectively. We highlight the possible performance degradation of the model after a hyperparameter tuning procedure when the weight factor distribution of the performance metrics is not carefully considered. We note several weight factors of interest that could serve as initial possible “safe spots” for further exploration. Full article

This work presents a time-domain analog-to-digital conversion method in which the amplitude of a sensor signal is encoded through its crossing instants with a periodic ramp. The proposed architecture departs from conventional ADC and PWM demodulation approaches by shifting quantization entirely to the time domain, enabling waveform reconstruction using only a ramp generator, an analog comparator, and a timer capture module. A theoretical framework is developed to formalize the voltage-to-time mapping, derive expressions for resolution and error, and identify the conditions ensuring monotonicity and single-crossing behavior. Simulation results demonstrate high-fidelity reconstruction for both periodic and non-periodic signals, including real photoplethysmographic (PPG) waveforms, with errors approaching the theoretical quantization limit. A hardware implementation on a PSoC 5LP microcontroller confirms the practicality of the method under realistic operating conditions. Despite ramp nonlinearity, comparator delay, and sensor noise, the system achieves effective resolutions above 12 bits using only native mixed-signal peripherals and no conventional ADC. These results show that accurate waveform reconstruction can be obtained from purely temporal information, positioning time-encoded sensing as a viable alternative to traditional amplitude-based conversion. The minimal analog front end, low power consumption, and scalability of timer-based processing highlight the potential of the proposed approach for embedded instrumentation, distributed sensor nodes, and biomedical monitoring applications. Full article

Excessive use of short-form video platforms such as TikTok has raised growing concerns about digital addiction and its impact on young users’ emotional well-being. This study examines the relationship between continuous TikTok exposure and emotional engagement in young adults aged 20–23 through a multimodal experimental design. The purpose of this research is to determine whether emotional engagement increases, remains stable, or declines during prolonged exposure and to assess the degree of correspondence between facially inferred engagement and physiological arousal. To achieve this, multimodal biometric data were collected using the iMotions platform, integrating galvanic skin response (GSR) sensors and facial expression analysis via Affectiva’s AFFDEX SDK 5.1. Engagement levels were binarized using a logistic transformation, and a binomial test was conducted. GSR analysis, merged with a 50 ms tolerance, revealed no significant differences in skin conductance between engaged and non-engaged states. Findings indicate that although TikTok elicits strong initial emotional engagement, engagement levels significantly decline over time, suggesting habituation and emotional fatigue. The results refine our understanding of how algorithm-driven, short-form content affects users’ affective responses and highlight the limitations of facial metrics as sole indicators of physiological arousal. Implications for theory include advancing multimodal models of emotional engagement that account for divergences between expressivity and autonomic activation. Implications for practice emphasize the need for ethical platform design and improved digital well-being interventions. The originality and value of this study lie in its controlled experimental approach that synchronizes facial and physiological signals, offering objective evidence of the temporal decay of emotional engagement during continuous TikTok use and underscoring the complexity of measuring affect in highly stimulating digital environments. Full article

Epileptic seizures arise from abnormally synchronized neural activity and remain a major global health challenge, affecting more than 50 million people worldwide. Despite advances in pharmacological interventions, a significant proportion of patients continue to experience uncontrolled seizures, underscoring the need for alternative neuromodulation strategies. Rhythmic neural entrainment has recently emerged as a promising mechanism for disrupting pathological synchrony, but most existing systems rely on complex analog electronics or high-power stimulation hardware. This study investigates a proof-of-concept digital custom-designed chip that generates a stable 6 Hz oscillation capable of imposing a stable rhythmic pattern onto digitized seizure-like EEG dynamics. Using a publicly available EEG seizure dataset, we extracted and averaged analog seizure waveforms, digitized them to emulate neural front-ends, and directly interfaced the digitized signals with digital output recordings acquired from the chip using a Saleae Logic analyser. The chip’s pulse train was resampled and low-pass-reconstructed to produce an analog 6 Hz waveform, allowing direct comparison between seizure morphology, its digitized representation, and the entrained output. Frequency-domain and time-domain analyses demonstrate that the chip imposes a narrow-band 6 Hz rhythm that overrides the broadband spectral profile of seizure activity. These results provide a proof-of-concept for low-power digital custom-designed entrainment as a potential pathway toward simplified, wearable neuromodulation device for future healthcare diagnostics. Full article