Jari Iinatti

Early detection of sepsis in Intensive Care Units (ICUs) is critical, as delayed intervention significantly increases mortality risk. This study investigates early sepsis prediction using multivariate physiological time-series data from the PhysioNet 2019 dataset, comprising 40 structured Electronic Health Record (EHR) variables recorded hourly. We evaluate lightweight deep learning models, including 1D-CNN, GRU, TCN, and a lightweight CNN–ViT hybrid, under a unified protocol targeting prediction up to 6 h before sepsis onset. This prediction horizon follows the PhysioNet 2019 benchmark and represents a clinically meaningful early-warning window that balances actionable intervention time with sufficient predictive signal. The results show that effective temporal modeling is key for early detection. The TCN achieves the most favorable balance between predictive performance and efficiency, with the highest recall (0.778), tied highest F1-score (0.495), highest AUPRC (0.50), and highest AUC (0.907 ± 0.010). It also records the lowest inference latency (0.62 s total over the test set under a CPU-only setting), supporting real-time applicability. In contrast, CNN–ViT achieves competitive performance (AUC 0.906 ± 0.011) but incurs substantially higher computational cost (44.34 s). Explainability using SHAP highlights clinically plausible indicators, including blood pressure, heart rate, oxygen saturation, and lactate. Overall, lightweight temporal models offer a practical and clinically meaningful solution for early sepsis prediction under real-time constraints.

Understanding radio-channel characteristics is essential for the design of reliable wireless brain computer interface (BCI) and brain-monitoring systems, where signal propagation is strongly influenced by the complex anatomy of the human head. In this paper, BCI radio-channel characteristics are obtained and analyzed using electromagnetic simulations based on multilayer human-tissue models and validated by measurements performed with a realistic 3D human head phantom. Radio channel analysis is performed considering several scenarios using a flexible implant and an on-body antenna. First, the effect of muscle tissue on channel attenuation is examined by comparing two tissue models with and without a temporalis muscle layer. The sensitivity of the channel parameters to relative antenna positioning is then analyzed by introducing small vertical displacements of the on-body antenna with respect to the implant antenna. Finally, the influence of subcutaneous fat thickness is investigated by considering typical anatomical variations as well as an extreme case representing severe obesity. These studies are done using two practical implant antenna locations, i.e., above the skull and within a shallow cranial recess. The results show that channel characteristics are strongly affected by tissue composition, implant depth, and antenna alignment, with these effects becoming more pronounced at higher frequencies. The findings highlight the importance of accounting for anatomical variability and antenna placement when designing reliable wireless BCI and brain-telemetry systems.

Sepsis is life-threatening organ dysfunction caused by a dysregulated host response to infection, making early detection critical for improving outcomes in intensive care units (ICUs). This study presents a retrospective comparative evaluation of deep learning architectures for predicting sepsis up to 6 h before the PhysioNet/Computing in Cardiology 2019 Challenge onset label using hourly structured electronic health record (EHR) variables, including vital signs, laboratory measurements, and demographics. Evaluated architectures include Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory (Bi-LSTM), Temporal Convolutional Network (TCN), Transformer, and hybrid Convolutional Neural Network–Vision Transformer (CNN-ViT) models. Median imputation and class-weighted loss were applied to address missing values and severe class imbalance, while Shapley Additive Explanations (SHAP) and attention analyses were used as complementary interpretability approaches. Among the evaluated models, CNN-ViT achieved the strongest overall minority-class performance, with 88.25% accuracy, 0.7480 recall, a 0.454 F1-score, and a 0.48 area under the precision–recall curve (AUPRC), although the numerical gains over other advanced temporal and hybrid architectures were modest. Leave-one-unit-out evaluation further demonstrated relatively stable performance under internal distribution shifts. The results suggest that combining local feature extraction with temporal and attention-based modeling can improve early sepsis prediction from structured ICU data. However, the study represents a retrospective computational benchmark using a public dataset and does not constitute prospective clinical validation or real-world deployment assessment.

Collaboration opportunities for devices are facilitated with Federated Learning (FL). Edge computing facilitates aggregation at edge and reduces latency. To deal with model poisoning attacks, model-based outlier detection mechanisms may not operate efficiently with hetereogenous models or in recognition of complex attacks. This paper fosters the defense line against model poisoning attack by exploiting device-level traffic analysis to anticipate the reliability of participants. FL is empowered with a topology mutation strategy, as a Moving Target Defence (MTD) strategy to dynamically change the participants in learning. Based on the adoption of recurrent neural networks for time-series analysis of traffic and a 6G wireless model, optimization framework for MTD strategy is given. A deep reinforcement mechanism is provided to optimize topology mutation in adaption with the anticipated Byzantine status of devices and the communication channel capabilities at devices. For a DDoS attack detection application and under Botnet attack at devices level, results illustrate acceptable malicious models exclusion and improvement in recognition time and accuracy.

Hybrid Wireless Body Area Networks (HyWBANs) hold significant promise for advancing in-body communication systems in healthcare. In this work, we propose a novel dual-modality framework that integrates Ultra Wideband (UWB) radio communications with Near-Infrared (NIR) optical wireless communication to enhance the performance of Wireless Body Area Networks (WBANs). UWB provides reliable deep-tissue connectivity, while NIR offers a low-power solution for shallow-layer data exchange under Line-of-Sight (LOS) conditions. We present extensive measurement campaigns using biological tissues and synthetic phantoms to characterize the performance of each modality independently, thereby establishing a comprehensive baseline for future integration. Our experimental results, which we have validated by using a propagation model, aim to provide the academic community with accurate data on which to build hybrid systems with increased communication reliability and adaptability under controlled conditions. Moreover, we discuss potential applications and outline decision criteria for dynamic mode selection based on real-time service requirements and environmental factors. This study lays the baseline for future research to develop a unified hybrid system that can adaptively optimize performance to meet the evolving demands of innovative healthcare.

This paper explores novel strategies to strengthen the security of Hybrid Wireless Body Area Networks (HyWBANs), which are essential in smart healthcare and Internet of Things (IoT) applications. Recognizing the vulnerability of HyWBAN to sophisticated cyber-attacks, we propose an innovative combination of semantic communications and jamming receivers. This dual-layered security mechanism protects against unauthorized access and data breaches, particularly in scenarios involving in-body to on-body communication channels. We conduct comprehensive laboratory measurements to understand hybrid (radio and optical) communication propagation through biological tissues. We utilize these insights to refine a dataset for training a Deep Learning (DL) model. These models, in turn, generate semantic concepts linked to cryptographic keys for enhanced data confidentiality and integrity using a jamming receiver. The proposed model significantly reduces energy consumption compared to traditional cryptographic methods, like Elliptic Curve Diffie-Hellman (ECDH), especially when supplemented with jamming. Our approach addresses the primary security concerns and sets the baseline for future secure biomedical communication systems advancements.

Wireless capsule endoscopy (WCE) is a minimally invasive procedure that allows for the examination of the gastrointestinal tract using a small, swallowable capsule equipped with a camera. Accurate localization of wireless capsule endoscopes within the gastrointestinal (GI) tract is pivotal for effective medical interventions. We propose a novel hybrid localization method, combining one-shot learning and trilateration, addressing challenges in the complex in-body wireless channel. Results indicate improved accuracy, surpassing traditional methods. Using a modified Laura model with increased organ sizes, a shift to a 4-zone configuration reveals altered electromagnetic interactions in the GI tract. This modification allows us to simulate scenarios involving larger individuals, providing insights into the adaptability of our proposed technique to diverse anatomical conditions. Despite the shift, our method excels, showing adaptability with superior performance in 4 and 5 zones across diverse locations. Sensitivity to environmental conditions is emphasized, impacted by factors like organ size. Similarity in results between 4 and 5 zones underscores adaptability, extending to 45° and 90° scenarios. In summary, our hybrid approach represents a promising advancement, enhancing accuracy for wireless capsule endoscope localization in varied GI tract environments.

The CENELEC TS 50701 is the first encompassing standard aiming at governing cybersecurity risk management processes within the railway industry. Although the technical maturity of this framework is undeniable, its application in practical projects is still an active field of discussion among practitioners, especially when dealing the communication-heavy subsystems. Among such subsystems, signaling is among the most critical ones. Both Communication-based Train Control (CBTC) and European Railway Traffic Management Systems (ERTMS) heavily rely on wireless communications for their operation. This paper describes two cybersecurity attack scenarios regarding wireless communications for CBTCs that can impact the safety of these systems using the lens of the framework provided by the novel CENELEC TS 50701. In doing so, we discuss the implications of using such guidance, especially concerning the different interpretations found in the literature regarding zoning communication systems, to assess and mitigate the cybersecurity risk and improve the posture of CBTC systems concerning the examined attacks. Experimental tests conducted in controlled laboratory environments and high fidelity simulations have been conducted to support the cybersecurity analysis.