Alzheimer's disease (AD) is described by progressive cognitive decline and neural network disintegration. Electroencephalography (EEG) phase synchronization, remarkably inter-trial coherence (ITC) offers a promising biomarker for evaluating functional connectivity deficits. However, its diagnostic utility during auditory gamma entrainment across different stages of AD remains underexplored. This study aimed to evaluate ITC as a neurophysiological marker of cognitive status by comparing EEG phase-locking responses to auditory gamma stimulation in cognitively normal (CN) individuals and patients with mild (mAD) and moderate Alzheimer' s disease (modAD). Twenty-seven elderly participants (9 per group: CN, mAD, and modAD) underwent EEG recording during resting state and during 40Hz auditory click-train stimulation. ITC was computed in the gamma band (30- 45 Hz) using Morlet wavelet decomposition. Group and condition level differences were assessed using repeated measures ANOVA. Classification accuracy was evaluated using support vector machines (SVMs) and receiver operating characteristic (ROC) analysis. ITC significantly increased during stimulation (p < 0.001, with the largest enhancement observed in CN, followed by mAD and modAD (p < 0.001). Phase locking was strongest in the frontal and temporal regions of CN participants. Classification accuracy reached 96.3%, with AUCs of 1.00 (CN), 0.95 (mAD) , and 0.92 (modAD). Auditory gamma stimulation induces strong phase synchronization in cognitively CN but not in AD patients. ITC is a sensitive, non-invasive biomarker of cognitive impairment and may support early diagnosis and therapeutic stratification in AD management.
Objective: Managing heart failure requires simultaneous regulation of mean arterial pressure, cardiac output, and left atrial pressure, which is challenging due to complex cardiovascular interactions and limited monitoring. This study presents a proof-of-concept in-vivo evaluation of a discrete two-degree-of-freedom hemodynamic control system designed to regulate these three hemodynamic variables through titration of cardiovascular interventions under pulmonary artery catheter monitoring. Results: Across five consecutive trials in three canine models of heart failure induced by tachycardia pacing, the cohort-averaged mean arterial pressure, cardiac output, and left atrial pressure were each within $\pm$10% of their respective targets when evaluated using the final three intervention measurements, with absolute percentage errors decreasing from 16.8% to 9.0%, 23.4% to 3.8%, and 87.5% to 5.5%, respectively. In a more severely complicated heart failure model with coronary embolization, outcomes revealed control limitation primarily due to pharmacological resistance, highlighting the importance of choosing the appropriate target value. Conclusions: This proof-of-concept in-vivo study demonstrates the feasibility of implementing discrete two-degree-of-freedom hemodynamic control under pulmonary artery catheter-based monitoring and phlebotomy-mediated volume removal in anesthetized canine models of tachycardia-pacing-induced heart failure. Further studies are required to evaluate the effects of inter-subject variability and clinical generalizability.
Cardiovascular diseases are the leading cause of global mortality, with millions of patients relying on implantable pacemakers. Finite battery energy fundamentally limits remote monitoring. Using open-source manufacturer longevity data we develop an energy-balance model to show a stringent ≈130 J/year communication budget, too low for continuous monitoring with conventional RF links. We investigate electro-quasistatic human body communication (EQS-HBC) as an alternative physical layer for pacemakers, combining a systematic review of implant-grade transceivers with a proof-of-concept channel feasibility study in a human-torso phantom. Our analyses indicate that EQS-HBC could, in principle, support 1 kSps, 16-bit equivalent continuous streaming at ≈ 2 μW (> 100× more energy-efficient than RF) while remaining within safety standards and preserving > 14 year lifetime, outlining design guidance for future implantable cardiac devices.
Goal: Contrary to conventional expectations, this work investigates the integration of Field Programmable Gate Arrays (FPGAs) in Implantable Medical Devices (IMDs), emphasizing their potential to improve computational performance under stringent energy and resource constraints. We advocate heterogeneous CPU-FPGA systems to support the growing demand for embedded AI in next-generation IMDs for intelligent and personalized healthcare. Methods: We evaluated heterogeneous CPU-FPGA architectures against CPU-only configurations across representative AI-driven neural and security IMD workloads. Results: FPGAs achieved lower execution latency and improved energy efficiency than CPU-based implementations in computationally intensive neural workloads, despite higher instantaneous power consumption in some cases due to parallel execution. In low-duty-cycle security workloads where aggressive power gating is feasible, FPGA- and CPU-based solutions exhibited comparable overall performance. Conclusions: These findings demonstrate that heterogeneous CPU-FPGA systems can integrate advanced AI and security algorithms into IMDs while satisfying battery-life, power-density, real-time processing, and practical implantability constraints relevant to long-term medical-device deployment, enabling intelligent, personalized, and secure implantable healthcare.
Goal: While Generative AI addresses data scarcity and privacy in healthcare, the lack of standardized validation protocols hinders the deployment of safe and trustworthy AI. This study introduces a comprehensive framework for rigorously evaluating the authenticity and fidelity of synthetic biosignals, addressing the critical need for medical data quality assurance. Methods: The framework was validated using synthetic electrocardiograms (ECG) for Brugada Syndrome generated via Generative Adversarial Networks (GANs). The protocol integrates four complementary analytics layers: (1) visual morphological inspection; (2) statistical assessment of clinical biomarkers (e.g., QTinterval); (3) computational similarity metrics (e.g. MSE, DTW) to assess distributional fidelity; and (4) a human-centered validation where blinded expert cardiologist scored clinical realism. Results: Results demonstrated high statistical fidelity and morphological realism. Quantitative analysis showed no statistically significant differences in key clinical features between real and synthetic distributions. An expert cardiologist achieved near 50% accuracy (random chance) in distinguishing real from synthetic signals, consistent with the high authenticity of the generated data. Conclusions: This framework proposes a comprehensive methodology for validating generative healthcare models, intended as an initial step toward standardization in the field. By ensuring high-fidelity data generation, this approach supports robust predictive modeling and AI-based decision support, facilitating privacy-sensitive clinical research.
Goal: A focused clinical hyperthermia (HT) system is presented for treatment of locally advanced breast cancer. Methods: The system consists of water-cooled phased array (PA) applicator integrated in treatment bed controlled by a programmable 434 MHz microwave source and patient-specific HT treatment planning (HTP) system with fiber-optic thermometry and user interface for system control and monitoring. Results: PA applicator performance in patient derived breast models and experiments in layered cylindrical phantom with HTP demonstrate the ability to deliver focused heating with low active reflection coefficients and channel power consumption. Conclusions: Thermal regulation and low cross coupling between PA elements significantly reduced the power consumption for focused HT, confirming that the proposed system is affordable and rugged for clinical adoption.
Glucose dysregulation in non-diabetic individuals can precede metabolic disease onset by years, yet most glucose prediction models target diabetic populations with structured insulin records. This article presents Patch Time-Aware Cross-Attention (Patch-TACA), a multimodal transformer-based framework for long-term blood glucose forecasting in healthy individuals. The approach combines patch-based self-supervised pretraining with a time-aware cross-attention mechanism to integrate continuous glucose monitoring (CGM) data with physiological and behavioral measures, such as accelerometry, heart rate, electrodermal activity, and meal macronutrients, without requiring explicit resampling of irregularly sampled signals. Patch-TACA is assessed using data from twelve healthy participants across three laboratory sessions. It achieves an RMSE of 14.26 $\pm$ 3.48 mg/dL at the 90-minute prediction horizon, outperforming GlySim and Gluformer baselines. Moreover, employing a self-supervised method for pre-training reduced RMSE by 7.9% compared to training with random weight initialization, and hyperglycemia prediction accuracy reached 93.9%. These results demonstrate that multimodal sensor fusion with self-supervised learning enables accurate long-horizon glucose forecasting in non-diabetic populations, supporting proactive metabolic health monitoring before clinical disease onset.
Goal: This work systematically evaluates a text-driven IMU data generation framework (Text2IMU) and identifies critical design choices to synthesise realistic wearable IMU signals as a basis for fully synthetically trained Human Activity Recognition (HAR). Methods: We synthesise wearable accelerometer and gyroscope signals from textual activity descriptions and evaluate the resulting data in a downstream HAR task. The IMU synthesis pipeline is systematically varied with respect to the number of textual prompts, surface models, and virtual sensor locations. In addition, we quantitatively compare synthetic data generated by different motion synthesis models as well as combinations of multiple models. Results: Across all evaluations, HAR performance varies substantially with the configuration of the Text2IMU pipeline. For correctly synthesised activities, the best combination yields a balanced accuracy of 89.2% when trained on synthetic-only data. Conclusion: The study demonstrates that diversity introduced through textual prompts, surface models, and virtual sensor placement is essential for accurate text-driven IMU data synthesis and fully synthetically trained HAR. Our results provide concrete guidance on suitable parameter ranges for Text2IMU synthesis within the evaluated configuration.
Early identification of motor impairment in infants at developmental risk can benefit from objective assessment methods. In this feasibility study, we evaluated quantitative kinematic metrics extracted from home-based recordings as indicators of developmental differences between twins with divergent perinatal outcomes. Two 4-month-old twins, one typically developing and one at-risk of motor impairment, were recorded using a single RGB-D camera at home. Videos of up to three minutes per twin were captured at seven timepoints, about 30 days apart. Seven upper body points of interest were tracked using DeepLabCut and their 3D coordinates calculated with an improved version of a previously published method adapted for home use. Thirteen metrics were estimated at each timepoint. Most metrics were consistent with the trends reported in the literature, distinguishing divergent perinatal outcomes. These findings demonstrated that the proposed method is feasible for home-based data collection, compensating for motion artifacts and reducing manual labeling time.
Goal: Objective: Renal Cell Carcinoma (RCC) is a heterogeneous cancer with multiple histological subtypes that require individualized clinical management. Once implemented, timely and appropriate subtype delineation from imaging data can lead to improved diagnostic accuracy and intervention decisions. Methods: Study compares 3 convolutional neural networks (CNN) architectures, AlexNet, GoogleNet, VGG16, Resnet18 and Densenet to determine which approach has the best performance for RCC subtype classification of Contrast-Enhanced computed tomography (CECT) images. The original dataset of 590 images from 90 patients was augmented using anatomically constrained augmentation (rotation ±5° and 2D scaling) to generate a total of 3469 unique images, with allocation of training, validation, and testing sets was 70:15:15, respectively. Each architecture was fine tuned to employ an optimized learning rate, dropout regularization, and L2 penalties to minimize overfitting. Results: The quantitative metrics for each architecture, displayed in Table IV, show that VGG16 achieved the maximum level of 99.21% accuracy but demonstrated some levels of overfitting behavior at the lower learning rates. GoogleNet achieved commendable accuracy levels of 97.67% accuracy after epoch optimization, but performance levels were highly sensitive across the learning rate. AlexNet, in comparison demonstrated the most stable convergence rates and achieved strong levels of generalization within all three large learning rates, achieving levels of 98.84% accuracy, and processing times were far lower than for either of the other networks. Conclusion: this work demonstrates that GoogleNet is a reliable and cost efficient implementation of a CNN for a clinical setting that enables rapid accurate clinical decision making from imaging data for the specific subtype of RCC. Significance: These findings reinforce the potential of tailored deep learning frameworks for early RCC subtype classification and encourage further validation on larger, multi-institutional datasets to enhance clinical applicability.
Objective: Predicting epileptic seizures using electroencephalogram (EEG) signals can provide a critical window for intervention to suppress impending seizures. However, the impact of varying interictal and preictal durations on seizure prediction remains unclear and warrants further exploration. Considering this, this study focuses on seizure prediction using intracranial EEG (iEEG) signals with variable interictal and preictal durations. Method: In this work, we discuss six pairwise combinations in total, involving two interictal durations (24 h and 48 h) and three preictal durations (30 min, 60 min and 90 min), during seizure prediction analysis. Furthermore, we also compare three classification strategies: the all-channel strategy (Strategy 1), the single-channel-based majority voting strategy (Strategy 2) and the channel-optimization-based majority voting strategy (Strategy 3), across each pairwise combination. Results: The long-term SWEC-ETHZ iEEG dataset is utilized to evaluate our method. The results reveal that our method achieves up to 87.7% sensitivity, 94.6% specificity, 92.8% accuracy, 0.912 Area Under the Curve (AUC) and 0.858 F1-score at the segment-based level, along with 95.9% sensitivity and 0.05/h false prediction rate (FPR) at the event-based level. Moreover, the overall performance of Strategy 3 surpasses that of Strategies 1 and 2 in each pairwise combination, indicating that channel optimization is beneficial for improving seizure prediction performance. Conclusion: This work shows high effectiveness in predicting seizures using iEEG signals, with the ability to forecast upcoming epileptic events up to 90 min in advance across both 24-h and 48-h interictal durations.
Goal: Existing beta burst detection algorithms for closed-loop deep brain stimulation (DBS) are computationally complex, limiting their use in implantable devices. We aimed to develop an improved beta burst extraction algorithm for chip-based DBS devices with real-time model updating. Methods: Building on an established beta burst detection method, we proposed a sliced mechanism for peak frequency finding and modified burst extraction for information sharing with real-time model updating. Results: Testing on rat electrocorticographic (ECoG) recordings showed that the proposed algorithm maintains a strong correlation ( 0.89 0.06) with the conventional method, with a 53.3 reduction in computational complexity for peak frequency finding. Conclusions: Integrating this improved beta burst detection into chip-based DBS devices represents a key algorithmic advancement toward adaptive neuromodulation therapies. The strong correlation and reduced complexity validate our proposal for real-time neural biomarker tracking, facilitating hardware and chip implementation, and advancing the development of implantable systems.
Tinnitus, the perception of sound without an external source, affects many individuals, yet its impact on the brain’s functional connectome remains underexplored. Traditional functional connectivity (FC) methods, such as Pearson correlation, phase lag index, and coherence, rely on pairwise comparisons between activity of macro-scale brain regions, limiting holistic characterization. We used an approach that estimates the entire connectivity structure by analyzing all time-courses simultaneously, robust even for short recordings and suitable for real-time applications. Using resting-state MEG from tinnitus patients and controls, learned connectomes outperformed correlation-based ones in fingerprinting individuals across test/retest. Group analyses revealed altered FC across multiple frequency bands, impacting default mode, auditory, visual, and salience networks, indicating large-scale reorganization. Tinnitus exhibited highly individualized whole-brain FC profiles, highlighting the importance of individual variability and paving the way for personalized models to optimize patient-specific interventions.
Goal: Accurate detection of malaria parasites using convolutional neural networks (CNNs) relies heavily on the quality of training annotations, yet creating quality annotations is both time-consuming and difficult to scale in high-burden, resource-limited settings. To address this challenge, we propose a method of annotating thin-smear blood images for the semantic segmentation of Plasmodium species, their developmental stages. Using a balanced collection of images from the Malaria Parasite Image Database, we trained identical SegNet models under three matched annotation regimes: expert manual labeling, SegNet-Only, and SegNet+Ontology where predictions are refined through biomedical ontological reasoning. Model performance was assessed not only for segmentation quality but also for how well each approach captured biologically meaningful information and for its interpretability as judged by clinicians. The proposed method produced results comparable to those achieved by expert annotations and clearly outperformed the baseline SegNet-only model in terms of biological consistency and clinical trustworthiness. The method successfully filtered out 5.7% of invalid AI-generated annotations by identifying semantic contradictions, ensuring the final training dataset adhered strictly to established biological constraints. Clinicians found the outputs from the proposed model nearly as reliable and understandable as those generated from expert annotations. These findings show that embedding formal biomedical knowledge into the annotation process can substantially reduce the cost and effort of creating training data while maintaining diagnostic accuracy and interpretability.
Noninvasive neuromodulation methods are promising for treating neurological diseases, but generally have limited ability to selectively target deep brain structures such as the basal ganglia or hippocampus. Transcranial temporal Interference Stimulation (tTIS) overcomes this limitation by superimposing high-frequency electric fields to create low-frequency amplitude-modulated envelopes that can preferentially target deep brain structures. This review explores the operation principles and recent advances of tTIS aimed at improving its spatial accuracy and versatility. Additionally, we analyze the key aspects of tTIS system design, including waveform generation, high-voltage-compliant output stage, and real-time charge balancing, which collectively enable safe and efficient stimulation delivery. We further discuss key challenges such as safety, standardization, and long-term efficacy, and outline future directions toward personalized tTIS.
Intra-body power transfer (IBPT) enables batteryless wearables by using the human body as a conductive medium. This work introduces Localized Capacitive Coupling (LCC), a new IBPT technique that uses a 40 MHz RF carrier to provide power transfer without relying on external grounds or environmental infrastructure. We evaluate LCC using computational modeling and human subject experiments with ten participants across multiple short-range capacitive body channels. Laboratory measurements with an isolated electrode system show mean path gains of 44 dB to 48 dB for channel lengths of 5 to 12 cm, closely matching our computational model with deviations under 3 dB. Circuit analysis indicates that air-gap coupling capacitance, typically in the femtofarad range, dominates channel gain, highlighting the importance of short-range fringing fields. To demonstrate practical energy harvesting, we designed and evaluated multistage Dickson charge pump (DCP) receivers. A five-stage impedance-matched DCP produced 3 V DC at 13.5 [Formula: see text] 0.5 dBm of incident RF power across a load in the tens of megaohms. A single-stage DCP paired with a battery manager generated a regulated 1.8 V output by charging a 100 [Formula: see text] capacitor from RF peak powers as low as 18.5 dBm. These results enabled a fully batteryless ring-worn motion sensor that uses an ultra-low-power accelerometer and non-volatile memory for offline activity logging, demonstrating that LCC is a practical approach for powering short-range wearable sensor networks. All hardware designs and simulation configurations will be open-sourced upon publication.
Goal: Analysis of the glottal area during vocal fold vibration has gained increasing attention. However, traditional analysis requires manual, frame-by-frame glottal area annotation to compute the glottal area waveform, a time-consuming, and error-prone process. Methods: This study proposes an automated system for glottal area segmentation and glottal area waveform feature extraction from 36 videostroboscopy recordings of 23 patients with vocal fold nodules. The system integrates YOLO and U-Net architecture for glottis detection and segmentation. Subject-independent 5-fold cross-validation was performed on 5017 annotated frames. Results: The system achieved an average Intersection over Union of 92.8%, and a Dice Similarity Coefficient of 95.8%, substantially outperforming thresholding and edge-based baselines.Computation time was reduced by 27.7 -folds compared with manual method. Applied to 14 patients undergoing voice treatment, the system detected consistent trends in glottal area dynamics post-treatment. Conclusion: The system enhances efficiency and accuracy of glottal area waveform analysis and demonstrates clinical utility for laryngeal assessment.
Goal: Skull fractures, especially those involving the cranial base and facial regions, present significant diagnostic challenges due to the skull's complex anatomy and subtle radiographic findings. Accurate detection requires repeated and meticulous examination of multiple CT slices, which is a significant cognitive burden, and requires considerable interpretation time. The primary objective of this study is to develop a visualization flattening technique that effectively transforms the curved skull surface into planar representations that enhance fracture features. Methods: A novel visualization process was developed that extracted the cranial surface and subsurface layers from head CT scans and used disk harmonic mapping to generate flattened representations of the lower, upper, occipital, and frontal hemispheres of the skull. The technique was applied to nine cases from the CQ500 dataset, with varying levels of inter-reader agreement, or lack thereof, among the original radiologists who interpreted the dataset. These cases encompass both straightforward and diagnostically challenging fractures that exemplify the advantages of the proposed methodology. Results: The flattened views unwrapped the fractures into continuous, high contrast features, with improved conspicuity compared to the fragmented appearance across multiplanar reconstruction slices. Comparison with existing skull visualization methods, the proposed technique demonstrated high contrast of fractures features, and delineation between emissary veins, with less distortion and high preservation of anatomical continuity. Conclusions: Disk harmonic flattening offers a new approach to skull fracture visualization, providing radiologists and emergency department staff with a valuable addition to the conventional radiological tools, particularly in diagnostically challenging cases.
Colorectal cancer (CRC) ranks third in incidence among all malignancies and is highly lethal in advanced stages. Combination chemotherapy regimens based on 5-fluorouracil (5-FU) remain the mainstay of colorectal cancer treatment alongside surgical resection. Even though new treatment modalities are emerging, many are either ineffective against KRAS-mutant tumors or prone to therapy resistance. Therefore, there is a critical need for new targeted therapies that may overcome the KRAS-driven chemoresistance and enhance the effect of chemotherapy. MicroRNAs can modulate several oncogenic pathways at once and can strengthen chemotherapy. In this study, we identified miR-873 as a potential chemosensitizer that modulates KRAS/MAPK signaling in CRC. We found that KRAS is overexpressed in metastatic versus primary tissues and in a large CRC patient cohort (n = 1,061), high KRAS expression was associated with worse overall survival (HR = 1.27; 95% CI, 1.04–1.56; log-rank p = 0.018). In vitro inhibition of KRAS by siRNA reduced clonogenic growth (HCT116, p = 0.0023; RKO, p = 0.0018) and invasion (p ≤ 0.0001). In silico prediction (TargetScan/miRWalk) analyses showed a conserved binding site between miR-873 and KRAS 3′UTR. Consistent with this prediction, miR-873 mimic transfection reduced KRAS protein expression and phenocopied KRAS knockdown by suppressing colony formation (p ≤ 0.0021) and invasion (p ≤ 0.0001) in KRAS-mutant HCT116 and KRAS-wild-type RKO cells. Dose-matrix screening and SynergyFinder+ analysis revealed synergistic inhibition of spheroid viability with miR-873 + 5-FU, including a low-dose pair (25 nM miR-873 + 12.5 μM 5-FU) showing positive synergy across ZIP/HSA/Bliss/Loewe models. In a poly(ethylene glycol)diacrylate(PEGDA) microwell 3D platform that generates uniform, size-controlled CRC spheroids, this combination produced the strongest suppression of spheroid expansion (day-5/day-3 area: HCT116, 0.61 ± 0.18 vs control, 2.08 ± 0.49; RKO, 0.66 ± 0.04 vs control, 2.08 ± 0.31) and reduced the live-cell fraction to ∼41% in both lines. Moreover, western blot analysis showed decreased KRAS and MAPK pathway activity (reduced p-ERK and context-dependent p-MEK), reduced Cyclin D1, and increased apoptotic readouts (cleaved PARP and a Bax/Bcl-2 shift). Together, these results position miR-873 treatment as a potential targeting approach to suppress KRAS/MAPK signaling and sensitize CRC to 5-FU and validate our PEGDA microwell 3D platform as a practical, translational testbed for miRNA–chemotherapy combinations.
Goal: In modern high-stress environments, effectively regulating cognitive arousal, through enhancement to boost engagement or inhibition to manage excessive stress, is essential for maintaining mental well-being and optimizing human performance. Hence, this study extends existing state-space models by integrating time-varying parameters and disturbance inputs for enhanced representation of arousal dynamics inferred from skin conductance. Methods: We augmented nominal models with time-varying parameters, then developed a recursive Bayesian estimator for state tracking. Simulation-based validation was performed using skin conductance data from six participants, drawn from an experimental dataset of noninvasive wrist-worn physiological recordings acquired during cognitive stress and relaxation tasks. Adaptive and robust control architectures were designed for closed-loop regulation of latent arousal states. Results: Simulations based on experimental data showed that both controllers outperformed static methods. On average, under inhibitory and excitatory conditions, the adaptive controller achieved average RMSE reductions of 26.9% and 51.6%, respectively, while the robust controller achieved reductions of 16.0% and 23.4%. In complex multi-step tracking, the adaptive controller reduced average RMSE by 33.7% and control effort by 18.5%; similarly, the robust controller reduced RMSE by 32.6% and control effort by 15.1%. Conclusion: These findings demonstrate that adaptive and robust control strategies can reliably manage dynamic arousal regulation, offering potential for real-world neuroadaptive systems supporting human performance and well-being.
