Objective: We present a novel RL driven approach for step width control using a robotic hip exoskeleton. Unlike reactive approaches that respond to balance perturbations, our design proactively and continuously adjusts step width with each step of walking so that users could be more resilient to perturbations. Methods: Hip exoskeleton was bilaterally operated via admittance control in abduction/adduction direction. The proposed control system incorporates a closed-loop linear feedback controller to adjust step width by modulating robotic hip joint stiffness. RL agent was designed to learn a policy to personalize the gains of the linear controller during walking. We employed a hybrid offline–online learning approach requiring only a single real-time learning episode up to 4 minutes to update the offline trained policy for each participant. The control framework was validated on 5 non-disabled and 3 participants with lower-limb amputations. Results: We showed that the system accurately achieved the desired step widths with an average error of 0.62 ± 0.48 cm across 24 targets and all participants. Further comparisons to a standard PID controller showed that our approach showed a statistically significant difference in step width error across trials (p<0.01) Conclusion: The controller demonstrated accurate step width control and robustness against user-induced step width disturbances. Significance: Given that step width plays an important role in human balance during walking, our hip exoskeleton and control framework could potentially advance proactive balance enhancement approaches.
Objective: To develop and evaluate a passive real-time MRI needle tracking framework for tip localization during in vivo robot-assisted needle insertions. Methods: Synthetic complex-valued MRI data representing needle-induced susceptibility artifacts across multiple orientations were used to train a 2D U-Net. During online operation, incoming gradient-echo images were converted into two-channel complex-valued inputs, segmented, and post-processed to extract tip coordinates. Orthogonal bi-plane imaging with dynamic positional scan plane updates provided high-resolution localization in all three directions. The framework was deployed online and evaluated during eight robot-assisted mock transperineal needle insertions in two porcine subjects. Results: The system achieved high per-scan tip detection success rates (SAG=98.3%, COR=94.8%) and per-cycle coordinate update success rates (3D=100%, full-resolution=93.0%). Mean 3D localization error relative to post-insertion confirmation MRI was 1.0$\pm$0.4 mm. Mean total processing time was 440$\pm$32 ms per scan, with a maximum of 542 ms. In a retrospective ablation analysis, a magnitude-only model trained on the same data showed substantially lower detection performance than the model using the proposed complex-valued input. Conclusion: Synthetic-data training enabled accurate in vivo passive needle tracking without real training annotations. Complex-valued input supported robust detection, while adaptive orthogonal scan plane updates maintained needle visibility and tip localization during insertion. Significance: This study provides the first demonstration of passive real-time MRI needle tracking with dynamic scan control during in vivo procedures and establishes a foundation for future closed-loop robot-assisted needle guidance under MRI.
Timely diagnosis of neonatal epilepsy is critical for mitigating long-term neurodevelopmental deficits. However, while neonates exhibiting extensive seizure burden are readily identified, accurately diagnosing epileptic infants with sparse or no overt seizure onset during monitoring remains a formidable challenge using electroencephalogram (EEG) signals alone. Current automated systems predominantly target ictal-specific seizure detection, limiting their clinical utility for risk stratification during seizure-free intervals (interictal), particularly given the pronounced EEG variability across neonates. This work presents a cross-subject deep learning framework that robustly distinguishes epileptic from “healthy” neonates by integrating dynamic graph topology with self-supervised adaptation. Our architecture models multichannel EEG as spatio-temporal graphs to capture functional connectivity, while a balanced test-time training mechanism dynamically adapts the model to unseen patients via masked signal reconstruction. This effectively mitigates the domain shifts caused by inter-subject variability without requiring patient-specific labels. Validated on a clinically curated cohort of 140 neonates (70 healthy, 70 epileptic), our approach achieves a segment-level accuracy of 83$\%$ and an area under the curve of 0.90. For the primary subject-level endpoint, accuracy increases from 0.79 without adaptation to 0.88 with balanced TTT, using aggregation thresholds selected on inner validation subjects and fixed before evaluation on each held-out fold. The cohort includes anoxic-ischemic encephalopathy, benign neonatal convulsions, and cerebrovascular accident. Analyses restricted to non-ictal windows also support discrimination between the two groups, although all neonates in the epilepsy group experienced seizures elsewhere in their recordings.
Objective: Regular blood pressure (BP) monitoring is needed to manage hypertension. However, cuff devices are not readily accessible, and many adults are living with uncontrolled BP. Furthermore, popular wearable cuffless devices typically require calibration with periodic cuff BP measurements or demographic information. A finger worn ring for cuffless BP measurement without such calibration was investigated. Methods: BP ring concepts were conceived based on the oscillometric cuff principle. The user varies their finger pressures similar to a cuff device by performing ring tightening followed by hand lowering or ring tightening only under visual guidance, while the ring obtains oscillometric measurements from the digital arteries for on-demand BP measurement. A prototype device was built to implement the BP ring concepts. The ring included a mechanical actuator to adjust its diameter and a photoplethysmography sensor, force sensor, and inertial measurement unit to make the measurements. The ring is interfaced to a display via a wired data acquisition system to guide the user in performing the ring tightening/hand lowering. An oscillometric algorithm is applied to compute BP from the measurements. The device was tested in 25 volunteers. Results: The ring tightening-hand lowering BP measurement could be learned by new users. The device yielded root-mean-square-errors of 7.1 and 6.9 mmHg for systolic and diastolic BP against a validated arm cuff device. Conclusion: This study demonstrates the feasibility of absolute BP monitoring using a finger worn ring. Significance: The BP ring concepts could potentially be incorporated in smart rings to help improve hypertension control.
Objective: Optimization of electrical stimulation patterns can enhance the performance of visual neuroprosthetics by replicating the selective and temporally precise neural responses characteristic of natural vision. Methods: An in silico approach employing the Artificial Bee Colony (ABC) algorithm was developed to optimize electrical stimulus waveforms that evoke physiologically realistic responses in retinal ganglion cells (RGCs). The optimization was guided by a computational model of ON and OFF RGCs, with the goal of matching recorded neural responses from the rat dorsolateral geniculate nucleus (dLGN). The ABC algorithm iteratively modified the amplitude envelope of charge-balanced, biphasic pulse trains to maximize similarity between simulated and recorded responses. Results: The ABC algorithm consistently achieved >= 80% cross-correlation with target neural responses within 2000 iterations. It also identified single optimized waveforms capable of eliciting distinct responses in neighboring neurons, indicating potential for simultaneous physiological responses with electrical stimulations. Under constrained computational conditions, ABC demonstrated superior accuracy compared to alternative optimization algorithms. Conclusion: The proposed ABC-based optimization approach effectively generates neural activation patterns closely resembling natural responses, supporting its use for real-time and adaptive stimulation control in visual neuroprosthetic systems. Significance: This study introduces a biologically inspired optimization framework that advances the development of intelligent stimulation strategies to enhance the functionality of next-generation visual neuroprostheses.
