This paper presents an initial investigation of the pulsed transmission strategy in optical wireless power transfer (OWPT) for medical implant devices (MIDs) case, aiming for deeper penetration across biological tissue. The main idea is to transmit high-power pulses for very short time periods, ensuring safe exposure through rapid switching periods while still attaining a reasonable average of harvested energy. This study considers an optical phantom to mimic human soft tissue. An 810 nm 375 mW NIR LED and a commercial indoor PV cell were used as transmitters and receivers in the OWPT testbeds, respectively. We compared the required time to charge the supercapacitor fully (storing approximately 5 J) and the energy harvested rate between the pulsed and continuous transmission methods over different levels of LED driving currents. Approximately 1.95 mJ/s in similar to 18 minutes of energy harvesting rate using 375 mW of transmitted power can be attained using pulsed mode (at 100 kHz). In contrast, similar to 4.41 mJ/s is attained in similar to 42 minutes using the same transmitted power at continuous mode. Results reveal that transmitting high illumination power across biological tissue in pulsed mode using readily available testbeds is feasible, albeit with a trade-off in terms of the supercapacitor's charging duration and the energy harvesting rate.
This paper focuses on an apnea/hypopnea as one of the major symptoms of Sleep Apnea Syndrome (SAS) and proposes its detection method based on knowledge distillation by reducing the gap between the contact and non-contact sensors. For this issue, the proposed method leverages the teacher model of the neural network trained by data of the high-quality contact sensors (e.g., the nasal pressure sensor or thoracic/abdominal belt sensor) when training the student model of the neural network with data of the low-quality non-contact sensors (e.g. the mattress sensor). To improve an effect of the above knowledge distillation, the following two data preprocessing methods are proposed: (i) the respiration amplitude normalization to suppress different amplitude caused by the different sleep positions; and (ii) the phase shift adjustment of the different respiration amplitude measured by the different sensors. To investigate the effectiveness of the proposed data preprocessing, this paper conducted the human subject experiment of the 35 subjects and revealed the following implications: (i) while the knowledge distillation without reducing the gap of the different sensors deteriorated the estimation, the knowledge distillation with the respiration amplitude normalization significantly improved the accuracy, precision, recall, and F1; (ii) an integration of the respiration amplitude normalization and its phase shift adjustment further and significantly improved the recall, which contributes not to overlooking the apnea/hypopnea.
This application-oriented study explores the potential of wireless power transfer (WPT) across biological tissues using commercial single-beam broadband near-infrared (NIR) LEDs, which is demonstrated on a testbed. We employ commercial indoor monocrystalline silicon photovoltaic (PV) cells as receivers in the testbed. Typically, PV cells are optimized for broad-spectrum light, such as sunlight or artificial light sources, which can cause inefficiencies when exposed to narrow-spectrum light like NIR due to spectral mismatches. To this end, a wide-spectrum (broadband) NIR LED is envisioned to enhance power conversion efficiency (PCE). To simulate biological tissues, tissue-mimicking optical phantoms were used. This preliminary study focuses on evaluating the open-circuit voltage (V-oc) of a commercial PV cell across various optical phantom thicknesses under different illumination levels of NIR light. The NIR LED used in this study emits in the wavelength range of 770-940 nm. The results show that PV cells can produce voltage from broadband NIR light penetrating optical phantoms up to 50 mm thick. This research emphasizes the potential of WPT using broadband NIR light for powering implantable electronic devices (IEDs).
Delirium is a common and serious condition among hospitalized patients, particularly in older adults, with significant clinical, social, and economic implications. Preventive measures targeting modifiable risk factors, such as cognitive stimulation and reorientation, have shown promise in reducing delirium incidence. This study explores the integration of manualized cognitive interventions delivered by volunteer medical students with continuous monitoring of motor activity using wearable sensors. The goal is to detect early signs of delirium and enhance prevention efforts. The project outlines a structured protocol focusing on hospitalized patients at risk of delirium in critical and non-critical care settings. Volunteer medical students serve as a bridge between healthcare professionals and patients, delivering daily cognitive interventions under supervision. Simultaneously, wearable sensors monitor patients’ motor activity, providing real-time data to identify deviations indicative of potential delirium onset.
Today, the early detection and prognostic assessment of primary melanoma of the skin are still significant challenges, as the disease remains a leading cause of mortality worldwide. Cutaneous melanoma prognosis is strictly related to accurate pathological Staging which includes primarily the measurement of Breslow tumor thickness, recorded to the nearest 0.1 mm, and tumor ulceration. Automatic classification of melanoma based on its Breslow thickness could improve patients stratification into clinically relevant groups. In this work, we present a custom Artificial Neural Network that classifies the severity stages by extracting asymmetry, border, diameter, color and textural features from melanoma dermoscopic images. Our model showed very high performance on both internal testing (91.41%) and external testing on a disjoint real-world image dataset (82.45%). Although future work should involve expanding the training and testing datasets, our model exhibited strong predictive and generalization capabilities.
It is well understood that sleep disturbances impact general quality of life. The Respeck device worn as a plaster on the chest was designed to monitor the respiratory signal respiratory rate (breaths/minute) and the respiratory flow/effort (amplitude), and the intensity of physical activity (in terms of equivalent volume of oxygen consumed (VO2) measure), and types of static and dynamic physical activity. Unsupervised machine learning methods considered for binary classification of sleep-wake states were k-means, hierarchical clustering, and anomaly detection applied to numerical features such as mean breathing rate, and activity levels derived from the Respeck data. K-Prototypes and Factor Analysis of Mixed Data (FAMD) which extended the k-means clustering with the activity type as a categorical feature, were the most effective in classifying the sleep-wake states. The FAMD with k-means method was applied to the SMILE dataset containing 904-days worth of Respeck data collected by a cohort of 17 patients with Chronic Obstructive Pulmonary Disease (COPD). For the first time night-time sleep quality features were quantified in COPD patients, such as sleep duration, sleep latency, sleep efficiency, wake-after-sleep-onset, and, number of positional changes using the Respeck sensor. It was estimated that 88% of the SMILE COPD cohort suffered from poor night-time sleep quality based on their satisfying threshold values for all three criteria: sleep duration, sleep efficiency and number of instances of wake-after-sleep-onset.
The overall reliability of the Internet of Things (IoT) enabled medical systems to rely on the captured data and associated services. Therefore, it is important to formulate non-malicious and unbiased data for data-driven medical IoT applications. In this paper, we identify the strength of blockchain and smart contracts in consensus-driven malicious data elimination for reliable medical data formulation for decentralized IoT medical services. Our work proposes a novel consensus protocol that leverages the reputation score as a numerical indicator to identify malicious data with privacy-preserved reputation verification for ensured fair and unbiased data formulation. We validated the proposed architecture by implementing an online medical diagnosis use case, and programmatic numerical event-driven simulations. The results reflect that our work outperforms the state-of-the-art by latency improvement (up to 57%) while supporting scalability and improved fairness(up to 97% node utilization in the experiment).
Gastrointestinal tumors that develop within the mucosal layer and remain undetectable by conventional endoscopic cameras have become a significant concern. Recently, the incorporation of radio channel analysis into capsule endoscopy has been proposed to facilitate the detection of tumors that are not visible to capsule cameras. This paper investigates the impact of abdominal adipose tissue thickness on the detectability of intestinal tumors in a radio channel analysis-based tumor detection system. The study is conducted using electromagnetic simulations to analyze channel characteristics between a capsule endoscope and on-body antennas with tissue models of varying thicknesses. Different tumor sizes and locations relative to the capsule are investigated. In addition to tissue layer models, human voxel models with different body constitutions are utilized. The results indicate that abdominal adipose tissue thickness has a clear impact on the detectability of tumors. However, changes in channel transfer functions are detectable even in the thickest evaluated cases if the tumor is located between the capsule and the on-body antenna. The changes observed in layer models are more pronounced than those in voxel models, which account for the realistic and complex propagation environment within the tissues. These findings offer valuable insights for the development of a channel analysis-based intestinal tumor detection system, which potentially will revolutionize the detection of tumors that are not visible to capsule cameras.
Digital pathology (DP), especially when implemented with artificial intelligence (AI) solutions, offers the promise of faster turnaround times, improved accuracy, and improved collaboration. However, these technological advances must be carefully balanced against the overall burdens they impose. In particular, the rapid change in the legal regulation of healthcare technology and the uncertainty about the applicable rules can be a major barrier to the development and diffusion of DP. The question is whether the spread of AI technologies in this field is an inevitable trend that should always be encouraged, or whether this transformation should be carefully managed because of its implications. In order to provide a primer answer and to highlight a research path, the analysis will focus on the diffusion of large language models (LLMs) in DP as a paradigm of the need to balance different public interests. The paper is the result of the collaboration between physicians, engineers and legal scholars.
Electronic Patient-Reported Outcome Measures (ePROMs) are widely used in telemonitoring for efficient patient status assessment, without the need of clinician intervention. However, the quality of collected data is often compromised by issues such as patient comprehension, response fatigue, and varying levels of digital proficiency. We’re looking at ways to overcome these challenges by using crowdsourcing techniques to evaluate the quality and reliability of patient responses. Our idea is to enhance PROMs by adding elements inspired by crowdsourcing, such as asking repeated or counterfactual questions, gauging self-reported confidence, and checking internal response consistency. This is all about improving how we understand and assess patient feedback without changing the original questionnaire’s structure. To show how this can work in practice, we’ve designed a proposal using the SNOT-22 questionnaire, which is used in otolaryngology to manage chronic upper airway diseases. Our plan involves adding extra questions and using metadata analysis to indirectly but effectively enhance response quality.
It is well-known, especially nowadays, how a regular amount of daily physical activity allows people to stay healthy, to reduce their risk of chronic diseases, and to improve their quality of life. To this end, technological solutions—in particular, wearable devices like smartwatches—can be considered as effective tools to enable self-monitoring of users, who can gain awareness of their health status and can possibly share information with medical personnel in the case of need. In this paper, an Internet of Things (IoT) architecture for long-time monitoring and daily physical activity quantification of healthy adults is presented. The proposed system exploits the availability of Garmin smartwatches worn by adult volunteers to collect multiple activity indicators, which are then properly processed in order to quantify the daily physical activity of each monitored subject. This is expedient to experimentally evaluate an innovative performance index, referred to as Physical Activity Index (PAI).
Understanding the health care providers’ perspectives is essential for successful deployment of artificial intelligence (AI) in health care. In this integrative review, we explored the health care providers’ perspectives about the use of AI for addressing mental health issues. We searched the MEDLINE, EMBASE, CINAHL, and PsycINFO from outset through November 2023. The following themes were identified from eight included studies: anticipating a forthcoming shift; AI acceptability; AI literacy; perceived capacity of AI for providing mental health care; potential benefits; and ethical and legal considerations. Findings showed that the providers’ perspectives about the AI capacity to deliver mental health care varied substantially across clinical tasks. Several apprehensions emerged regarding ethical, legal, and regulatory aspects of the technology use. Clarification of regulatory and ethical issues, transparency in model development, improved AI literacy, and involvement of all relevant stakeholders in development process can facilitate realistic, receptive, and safe deployment of the technology in clinical settings.
Stroke is a leading cause of disability and mortality worldwide, necessitating early detection for effective intervention. This study introduces a novel, mobile-enabled solution for early stroke detection, leveraging a lightweight deep learning (DL) approach to identify acute and non-acute stroke symptoms from facial features in real time. The proposed system utilizes the YOLOv8n model, a state-of-the-art object detection architecture, which has been fine-tuned on a custom dataset tailored for stroke-related facial anomalies. To ensure compatibility with resource-constrained devices, the trained YOLOv8n model was converted to TensorFlow Lite, a framework optimized for mobile deployment. The system is integrated into an Android mobile application using Flutter, a cross-platform development framework, enabling seamless execution and real-time video streaming from the device's camera. This cutting-edge implementation allows for continuous health monitoring, providing users with immediate feedback on potential stroke symptoms. The lightweight nature of the TensorFlow Lite model ensures efficient performance on mobile devices without compromising accuracy. Experimental results demonstrate the system's ability to detect stroke-related facial asymmetries and anomalies with high precision, making it a promising tool for early diagnosis and timely medical intervention. By combining advanced DL techniques with mobile technology, this work paves the way for accessible, real-time health monitoring solutions, particularly in remote or underserved areas where immediate medical attention is often unavailable.
Traditional medicine has advanced significantly in the recent years, but some pathologies remain difficult to treat. Implantable medical devices (IMDs) offer a promising solution, providing continuous monitoring, early detection, and targeted treatment for dysfunctional organs, complementing traditional therapies. Preclinical studies are essential to validate IMD safety, effectiveness, and interactions with traditional approaches. However, powering IMDs is a major challenge. This study presents a Wireless Power Transfer (WPT) technique based on inductive coupling to power a chip implanted in a freely moving rat within a cage. We evaluate various coil geometries to optimize WPT efficiency and minimize heating. Using MATLAB and COMSOL software, we analyze two-and multi-coil configurations to ensure continuous power delivery while meeting safety standards for SAR and temperature. Our results show that geometric optimization enhances WPT, reduces energy waste, and minimizes radiation exposure.
Voice communication is essential to effective exchange and sharing of instructions and patient information in clinical settings, with most communication between inpatients and staff (usually nurses) done verbally person-to-person or through the nurse call system. In this paper, we describe the current status of mobile voice communication systems, which have become indispensable to modern hospitals. We also discuss potential alternatives to the PHS systems that have been widely used in Japan but are being phased out, and show the features of four post-PHS candidates that can be used as a nurse call system. All systems have their advantages and disadvantages, but systems that function in the dedicated frequency band of 1.9 GHz, which is currently operating stably in hospitals, will have advantages. We recommend sXGP from the viewpoint of technology and minimal impact on other systems.
Rapid integration of wireless technologies into healthcare is revolutionizing the way medical data is collected, transmitted, and analyzed. This paper addresses emerging synergies between the Internet of Medical Things (IoMT) and telemedicine, emphasizing the transformative potential of Wireless Body Area Networks (WBAN) and related protocols. The study will help to understand the critical role of Bluetooth Low Energy (BLE) and ETSI SmartBAN in enabling low-power, secure communication for wearable devices and health monitoring systems. Using Texas Instrument (TI) LAUNCHXL-CC2650 devices, key performance metrics, such as current consumption, throughput, and packet error rate (PER) are analyzed under real-world conditions and this research uncovers the strengths and limitations of BLE, paving the way for a comprehensive comparison with the SmartBAN protocol. The findings not only bridge gaps in the current literature but also establish a foundation to improve SmartBAN protocol designs, aligning them with real-world healthcare needs. By focusing on the interplay between theoretical and practical aspects of WBANs, this work fosters innovation in IoMT, laying the groundwork for sustainable, efficient, and personalized healthcare solutions.
Accurate Electrocardiogram (ECG) classification is crucial for real-time cardiac monitoring. This study integrates static and wavelet-based scattering transform features to classify four heart rhythms: normal (N), other (O), atrial fibrillation (A), and noisy signals (similar to). Extracted features include mean, standard deviation, root mean square, skewness, kurtosis, and band power. The Symlet-2 wavelet, chosen for its symmetry, enhances classification accuracy by optimizing decomposition levels to balance feature retention and noise reduction. Zero-padding or symmetric padding mitigates boundary effects, preserving feature integrity. We evaluate LSTM, CNN-LSTM, TCN, and Transformer models using 5-fold cross-validation on the imbalanced PhysioNet 2017 dataset, employing F1-score and AUC-ROC metrics. Feature extraction improves accuracy, with CNN-LSTM achieving 66.83% accuracy, an F1-score of 0.67, and an AUC of 0.75-0.82. LSTM and TCN show moderate performance (F1-score 0.59, AUC 0.52-0.72). Without feature extraction, all models perform worse, though CNN-LSTM remains the best (AUC 0.67-0.83). Feature extraction also reduces inference time from 274.40s to 155.94s. Data augmentation (time warping, jittering, time masking, magnitude warping, scaling, cropping) further improves CNN-LSTM performance for Classes N, A, and O, though Class similar to remains unaffected. These results underscore the effectiveness of hybrid feature extraction and augmentation in enhancing ECG classification for real-time health monitoring.
Microwave imaging has gained attention for its potential in medical applications, particularly breast cancer imaging. The use of safe, non-ionizing radiofrequency signals makes it a promising alternative for early detection in screening programs. This study presents preliminary findings from a clinical investigation in one of the centers using the microwave imaging MammoWave (R) device, enhanced by a machine learning-based clinical decision support system (CDSS). Conducted across nine European hospitals, the clinical study (ClinicalTrials.gov: NCT06291896) aims to validate MammoWave's sensitivity (>75%) and specificity (>90%) using data from 10,000 volunteers. Initial results from 760 volunteers indicate high compliance and promising lesion detection capabilities, although additional efforts should be made to achieve balanced sensitivity and specificity. As this is a work-in-progress study, ongoing data collection and analysis will be crucial to refining MammoWave's clinical performance. By sharing these preliminary results, we aim to foster discussion, gather critical feedback, and explore potential refinements in microwave-based imaging.
Body Area Networks (BANs) are part of constrained networks. Due to the limited power consumption and processing capabilities of wearables and special medical implants, conventional cybersecurity protocols based on digital certificates may not be the best choice due to the processing involved. Moreover, access to a Public Key Infrastructure may not be available at all times or at the required time in a BAN. Nevertheless, secure communication links in BANs must always be operating. The proposed solution assumes access to the BAN device's manufacturer repository only for the one-time mutual authentication of BAN devices (wearables, medical implants, coordinators) the first time such devices are used by a user. After successful one-time authentication of the cryptographic credentials, the link to the manufacturer repository is no longer needed, and BAN devices run a variation of an Asymmetric Password Authentication Key Exchange for registration of credentials in the BAN coordinator and request to associate to a secured BAN at any time without digital certificates and access to a Public Key Infrastructure.
Parkinson’s disease (PD) is a progressive neurodegenerative disorder that significantly affects motor function. In clinical practice, symptom severity is typically assessed through subjective evaluation. The Movement Disorder Society - Unified Parkinson’s Disease Rating Scale includes the Leg Agility test to assess lower limb bradykinesia, traditionally evaluated through visual observation. Objective evaluation methods are still not commonly used. To enhance accuracy and reproducibility, this work proposes an inertial measurement unit (IMU)-based algorithm for estimating the amplitude during the Leg Agility test. Our method processes acceleration and angular velocity signals from IMUs placed on the foot and thigh and compares them to determine optimal sensor placement. Validation was conducted using a motion capture system as ground truth. The results demonstrate that the thigh-mounted sensor yields superior accuracy, with a median root mean square error of 1.08 cm for the thigh and 1.84 cm for the foot, and a mean absolute peak error of 0.96 cm and 2 cm, respectively. Our algorithm shows good performance in capturing amplitude trends, a key factor in clinical assessment of the Leg Agility test. These findings support the feasibility of IMU-based monitoring for objective PD symptom assessment, offering a practical and objective alternative to traditional methods.
