Objective: To develop and evaluate an AIoT-enabled stereoscopic virtual-reality (VR) three-dimensional (3D) mental-rotation (MR) assessment, integrated with a rehabilitation robot, for objective measurement of cognitive impairment (CI). Methods and Procedures: Seventeen participants—eight older adults with clinically diagnosed mild cognitive impairment or Alzheimer’s disease (mean age 74.3) and nine healthy controls (mean age 50.1)—completed a 12-trial interactive stereoscopic 3D MR task. Task outcomes and kinematic behaviors (movement distance/velocity, rotation amount/speed; 192 features) were transmitted to an AWS-based AIoT platform and analysed with trial-aggregated two-sample t-tests under false discovery rate (FDR) correction with Hedges’ g effect sizes, and with participant-level supervised classification (k-NN, SVM, RF, DT, LR) under leave-one-out cross-validation. Results: Of 192 features, 50 differed between groups before correction; aggregating each measure across trials and applying FDR correction, eight of 16 kinematic measures separated the groups with large effect sizes (|g|= 1.18–1.47). Total rotation did not differ between groups, whereas rotation rate was less than half in the impaired group (g = −1.27), identifying the speed rather than the amount of visuospatial manipulation as the characteristic that most separated the groups. Participant-level classification reached an AUC of up to 0.76 and is reported as a preliminary feasibility benchmark. Conclusion: The framework reliably instruments a stereoscopic mental-rotation task and identifies rotation rate as the measure that most separated the groups. As the groups were not age-matched, the respective contributions of cognitive impairment and age-related slowing will be established in a larger, age-matched cohort, which is the next step toward clinical use. Clinical Impact: This study presents an AIoT-enabled stereoscopic mental-rotation assessment for objective evaluation of cognitive impairment in older adults. By integrating VR task performance, robot-derived movement signals, and cloud-based analytics, the system produces quantitative measures of visuospatial action without subjective scoring. The analysis identifies the rate of object manipulation as the characteristic that most separated the impaired group from controls in this feasibility cohort. Clinical and Translational Impact Statement: Early/Pre-Clinical Research: Instrumented stereoscopic mental rotation yields objective kinematic measures that separate older adults with and without cognitive impairment, with rotation rate emerging as the most informative marker. Delivering assessment at the point of contact while analysing it centrally suits community screening programmes where specialist time is limited. Validation in an age-matched cohort against established instruments is required before clinical adoption.
Despite major advances in reperfusion strategies, acute ischemic stroke remains characterized by limited therapeutic windows and a persistent need for neuroprotective interventions acting beyond revascularization alone. Following promising preclinical and early clinical experiences suggesting that Pulsed Electromagnetic fields (PEMFs) promote neuroprotection, a double-blind, placebo-controlled, randomized study (I-NIC study, NCT02767778) was designed and showed that PEMFs exposure, applied within 48 hours of stroke on-set, could indeed reduce cerebral lesion volume. Given these positive clinical results, a patient-level quantitative assessment linking delivered treatment to lesion evolution in such controlled clinical setting is required. Hence, in the present work we characterized how acute ischemic lesions responded to PEMFs treatment by performing individualized computational dosimetry. The semi-specific modeling approach, that combines individualized segmented lesions and coil placement with a generic head model, was applied to replicate the treatment administered to each enrolled patient and to estimate the resulting magnetic flux density (B-field) distribution. Individual responses were quantified through a distance resolved formulation of the exposure to PEMFs, where the lesion distance from the coil acted as a grounded coordinate that linked its response to the B-field attenuation, so deriving a spatial dose–response relationship. Through this analysis, greater lesion reduction in active patients was observed predominantly in regions closer to the stimulation coil, consistent with a spatially localized, field-intensity-dependent effect. Therapeutic effectiveness, defined as the achievement of a 50% volume reduction, could be reached in lesion regions placed within 47 mm from the coil, corresponding to a local B-field intensity of 1.3 mT. Together, these results provide a quantitative interpretation of the clinical response observed in the I-NIC study and indicate how coil position and spatial relationship to the lesion influence its exposure, offering practical elements for optimizing stimulation parameters and informing future patient-specific approaches.
Objective: Frailty screening is important for early risk stratification and care planning in older adults, but conventional assessments are often resource-intensive and difficult to scale beyond clinical settings. Methods: This study proposes a context-aware webcam-based Artificial Intelligence of Medical Things framework for four-class frailty screening using skeleton–environment fusion. Frontal walking videos were captured using a standard RGB webcam. MediaPipe-based pose estimation was used to extract full-body skeletal sequences, while silhouette-based subject removal was applied to reduce privacy-sensitive visual information before environmental feature extraction. Temporal gait features were learned using an LSTM network, and environmental context was extracted using an ImageNet-pretrained VGG16 model. The two feature streams were fused for Clinical Frailty Scale-based classification. Results: The skeleton-only model achieved an F1-score of 0.73. The proposed skeleton–environment fusion model improved the F1-score to 0.93, with an accuracy of 0.93, precision of 0.94, and recall of 0.93. Conclusion: Environmental context provides clinically meaningful information for webcam-based frailty screening, particularly when assistive devices alter gait patterns. The proposed framework supports scalable, low-cost, contactless, and privacy-aware frailty monitoring.
Background Intermittent blood-pressure (BP) and hemodialysis (HD) machine measurements limit continuously updated assessment of the intradialytic hypotension (IDH) risk. We developed and internally validated a multimodal machine-learning framework integrating noncontact radar-derived vital-sign summaries with HD machine parameters and demographics for 10-min-ahead IDH prediction. Methods This single-center study included 233 HD patients, 6147 sessions, and 132,608 observation windows. Among these, 375 windows (0.28%) preceded clinically documented IDH within 10 min, corresponding to 325 episodes in 46 patients. Models were evaluated using patient-wise nested five-fold cross-validation. The primary early-fusion model combined demographic, HD machine, and radar-derived heart- and respiratory-rate features. Regression and permutation feature importance (PFI) analyses were performed to evaluate the clinical associations and identify key contributors. Results The multimodal early-fusion model achieved an area under the receiver operating characteristic curve (AUROC) of 86.62% (95% confidence interval, 82.19%–89.49%), area under the precision-recall curve (AUPRC) of 3.33% (2.23%–6.09%), recall of 78.72% (67.81%–85.31%), specificity of 78.73% (72.90%–83.37%), and precision of 1.09% (0.71%–1.44%). It detected 260 of 325 IDH episodes (80.00%) with 4.53 false-positive prediction windows per session. A clinical baseline using recent systolic blood pressure and ultrafiltration-rate information achieved a higher AUPRC (4.21%). HD machine features accounted for most predictive information; incremental contributions from radar and demographics were modest and statistically uncertain. Conclusions Multimodal 10-min-ahead IDH risk estimation was feasible under internal validation, although most short-term predictive information was derived from HD machine parameters. Further work is needed to establish the incremental clinical value of continuous radar-derived monitoring.
Objective—Low-burden monitoring of pediatric fine-motor rehabilitation remains difficult because standardized clinical assessments are episodic, examiner-dependent, and rarely integrated with repeated training. This study presents a research-use webcam-based serious-game system intended to support therapist-supervised fine-motor assessment and longitudinal progress monitoring in children with developmental delay (DD). Methods—The system integrates MediaPipe hand tracking with two occupational-therapy-informed games: FingerBubble for distal finger dexterity and Gacha Machine for proximal wrist rotation and stability. Kinematic descriptors of trajectory variability, acceleration control, range of motion, and rotational velocity were extracted from RGB video and analyzed using participant-level machine learning. A typically developing cohort (N = 21) provided normative comparison data, and children with DD (N = 7) completed a 12-week, 24-session intervention protocol under therapist supervision. Results—The best hybrid model integrating distal and proximal indicators achieved a leave-one-subject-out accuracy of 0.821 and an F1-score of 0.615 for baseline DD/TD separation. Longitudinal analysis showed a significant reduction in distal trajectory variability, together with an expansion of wrist range of motion and a reduction in rotational velocity that did not reach significance at this sample size, and a leftward shift of model-predicted DD probabilities toward the typically developing range after training. Conclusion—The proposed framework demonstrates the feasibility of interpretable, low-cost, vision-based digital biomarkers for therapist-supervised pediatric fine-motor rehabilitation monitoring. The results support feasibility and measurement validity, but they should not yet be interpreted as definitive clinical efficacy evidence or autonomous diagnosis. Clinical and Translational Impact Statement—This work is positioned as a translational-enabling prototype at Technology Readiness Level 5. The system has been evaluated in a clinically relevant pediatric rehabilitation environment, but it is not integrated into routine clinical workflow and has not obtained medical-device regulatory approval. The revised manuscript therefore defines the intended use as research-use clinical decision support, clarifies safety and therapist oversight, and specifies future requirements for multi-site validation, workflow integration, regulatory planning, and AI/ML reporting before clinical deployment.
