Marek Penhaker

This study investigates how different controlled respiration patterns influence the coordinated dynamics of facial muscle activity and cardiac function by analyzing electromyography (EMG) and electrocardiography (ECG) signals. Twenty-six healthy participants performed four respiration tasks with different controlled respiration conditions, while facial EMG and ECG signals were continuously recorded. To characterize the temporal structure of these physiological signals, three nonlinear complexity metrics — fractal dimension (FD), sample entropy (SampEn), and approximate entropy (ApEn) — were computed from EMG signals and heart rate variability (HRV) derived from ECG. The results revealed that the complexity of both EMG and HRV signals increased as the difficulty of the respiration tasks increased, indicating more irregular and information-rich physiological dynamics under more demanding breathing patterns. In addition, strong positive correlations were observed between the complexity variations of facial EMG and HRV across the respiration conditions, suggesting coordinated adaptation between somatic and autonomic physiological systems during controlled breathing. These findings demonstrate that nonlinear complexity measures provide sensitive indicators of cardio-somatic responses to respiration modulation and highlight the potential of combined EMG–ECG complexity analysis for understanding physiological regulation during breathing tasks and related psychophysiological states.

Objective To develop and experimentally validate a finite element (FE) model of a clinical cardiac pacemaker pulse generator casing and position the model within contemporary device-mechanics and model-credibility expectations. Methods A commercially used Guidant Contak Renewal TR-2 pacemaker was destructively analysed to obtain the geometry of the titanium housing, internal lithium–polymer battery and plastic supports. Housing composition was verified by emission spectroscopy and elastic parameters were taken from tabulated data, while an effective Young’s modulus of the encapsulated battery was derived from quasi-static compression testing, consistent with current homogenized component-modelling practice. The three-dimensional FE model was built in MSC Patran/MD Nastran using shell elements for the thin-walled housing and ribs and solid elements for the battery. Boundary conditions reproduced a quasi-static compression experiment with bonded steel loading plates up to a force level associated with temporary loss of function. Results The validated model reproduced the measured force–deflection response with a relative deviation of 6–15% at representative load levels. Displacement and von Mises stress fields identified mechanically critical regions, particularly near curvature transitions, welds and housing–battery contacts. Conclusions The validated FE model provides a mechanically realistic baseline for casing behaviour under external compression and supports design optimisation, safety-margin assessment and targeted mechanical test development, with explicit credibility limits aligned to its intended context of use.

Over four decades of retinal vessel segmentation (RVS) research (1982–2025) are synthesized, tracing developments across classical image processing, traditional machine learning, and deep learning paradigms. A cross‐paradigm taxonomy is introduced that links computational methods with clinical interpretability and model generalization, addressing gaps identified in earlier surveys. A comparative meta‐analysis of 428 studies catalogs preprocessing and postprocessing techniques, segmentation architectures, and loss functions and assesses reported performance on benchmark datasets (DRIVE, STARE, CHASE_DB1, HRF) using 20 standardized evaluation metrics. Architectural trends, including residual‐enhanced U‐Nets, attention‐guided models, graph convolutional networks, and transformer‐based architectures (SGAT‐Net, MTPA‐UNet), correlate with measurable improvements in vessel continuity, topology preservation, and clinical reliability. Emerging interpretability strategies, including attention mapping, saliency visualization, and feature attribution, are examined for their ability to enhance clinical transparency and trust in automated RVS systems. Persistent challenges include cross‐dataset generalization, annotation scarcity, and limited clinical deployment of explainable models. A forward‐looking roadmap emphasizes domain adaptation, semi‐supervised and weakly‐supervised learning, and efficient architectures for clinical translation. By integrating quantitative evidence with a clinically grounded taxonomy, a novel perspective on RVS is provided, serving as a comprehensive reference for research and clinical adoption in precision ophthalmology.

Major depressive disorder remains a critical psychiatric condition, necessitating objective diagnostic instruments beyond conventional self-reported assessments. Electroencephalogram (EEG)-based deep learning (DL) approaches provide a non-invasive and data-driven solution. Yet, prior models have been limited in generalizability due to evaluation on a narrow set of datasets, and their lack of interpretability restricts clinical applicability. This paper presents AttResNet, a compact and efficient DL-based measurement instrument for electrical EEG signal processing and classification. In contrast to earlier approaches, which focus on algorithmic accuracy, AttResNet is developed as a measurement system optimized for embedded platforms. The model exhibits low computational complexity with 2.02 M parameters, requiring only 8 MB of storage, and is successfully implemented on a raspberry pi 4, where it demonstrates an inference time of 0.0059 s. Rigorous evaluations on three independent EEG datasets validate its robustness, achieving mean subject-independent classification accuracies of 99.7%, 96.7%, and 99.9%, surpassing existing approaches while ensuring generalizability. AttResNet enhances the design and development of biomedical instrumentation for mental health monitoring by combining modules for optimal signal measurement, feature extraction, and decision-making. Its implementation on embedded systems enables real-time EEG-based depression detection, marking a significant step toward scalable, wearable, and accessible instrumentation for mental health diagnostics in the internet of things era.

Pancreatic tumor segmentation from computed tomography (CT) images remains a challenging task due to limited annotated datasets, pronounced anatomical variability, and the high computational demands of state-of-the-art deep learning models, which collectively hinder their routine clinical adoption. This study proposes a clinically oriented end-to-end framework that bridges methodological advances in deep learning with practical deployment by enabling adaptive segmentation under realistic data growth scenarios. Rather than introducing a novel segmentation architecture, the framework integrates existing convolutional and transformer-based models within a lightweight graphical user interface (GUI) and employs a recursive augmentation strategy as a simulation mechanism to emulate the incremental availability of annotated clinical data over time. Multiple candidate architectures were first evaluated using cross-validation, after which representative lightweight and high-capacity models were selected for recursive augmentation. The framework was subsequently evaluated using both CNN-based architectures, such as 3D U-Net, and transformer-based models, such as VT-UNet-B, on multiple large-scale public datasets. Across all experiments, the proposed recursive augmentation consistently improved segmentation performance relative to baseline training, yielding relative Dice Similarity Coefficient (DSC) gains in the range of approximately 4–11% before reaching architecture-dependent saturation. Lightweight CNNs exhibited earlier saturation with smaller but consistent improvements, whereas transformer-based models benefited more substantially from incremental data expansion. By embedding segmentation models into an interactive GUI that supports real-time visualization and expert-driven refinement, the proposed framework emphasizes deployment feasibility, adaptability, and continuous performance improvement. The results outline a practical pre-clinical pathway toward resource-aware pancreatic tumor segmentation in real-world healthcare environments.

Background: The objective assessment of hearing in non-cooperative populations, such as neonates, remains a challenge. While Brainstem Evoked Response Audiometry (BERA) is the gold standard, its sensitivity to motion artifacts necessitates alternatives. Objective: This scoping review maps the current literature on functional near-infrared spectroscopy (fNIRS) as a supplementary method in objective audiometry. Data Synthesis: fNIRS shows potential to detect cortical hemodynamic responses, particularly to complex stimuli like speech, which BERA cannot fully assess. Key advantages include motion tolerance and suitability for pediatric and cochlear implant populations. However, the literature reveals significant heterogeneity in stimulation protocols and data processing. Evidence suggests fNIRS is better suited for assessing higher-level auditory processing rather than replacing BERA for threshold estimation. Conclusions: fNIRS is a promising complementary tool. However, due to the lack of standardized protocols and large-scale validation studies, it is not yet a direct clinical replacement for BERA. Future work must focus on protocol standardization and establishing robust normative data.

Background and motivation: Liver segmentation plays a critical role in medical imaging, aiding in diagnosis, treatment planning, and surgical interventions for liver diseases. Precise segmentation of liver structures, including vessels, tumors, and other substructures, is essential for effective patient management. Traditional manual methods are time-consuming and prone to variability, prompting the development of automated techniques. This review aims to evaluate the evolution of liver segmentation methodologies, focusing on recent advancements in deep learning and hybrid approaches. Materials and methods: This review follows the PRISMA guidelines for systematic analysis, including a detailed database search across PubMed, Web of Science, Scopus, and IEEE Xplore. The search focused on segmentation techniques for various liver structures using deep learning, traditional methods, and hybrid models. A total of 7819 studies were initially identified, with 190 selected for detailed analysis based on inclusion criteria like Dice Similarity Coefficient (DSC) metrics and clinical applicability. Results: The analysis identified deep learning models, such as U-Net variants and Swin Transformer-based architectures, as leading methods for liver parenchyma and tumor segmentation, achieving DSC values up to 98.9% on benchmark datasets. For vessel segmentation, methods like DeepLabV3+ and the feature-based approaches demonstrated robustness across different datasets. Despite progress, challenges remain in segmenting structures like biliary ducts and hematomas due to limited annotated data and imaging variability. Discussion: While deep learning has significantly improved segmentation accuracy, challenges such as class imbalance and variability across imaging modalities persist. Hybrid approaches that combine traditional image processing with advanced neural networks show potential for further improvement. Future research should focus on enhancing generalizability through multi-modal data integration and exploring semi-supervised learning methods to overcome data scarcity. Conclusion: This comprehensive review highlights the advancements and ongoing challenges in liver segmentation, emphasizing the need for continuous innovation. By addressing current limitations, future methodologies can improve accuracy, efficiency, and clinical relevance, ultimately enhancing patient outcomes in hepatology.