Journal: Scientific Reports

Article Title: Lightweight and interpretable edge intelligence AI with intrusion detection for trustworthy cardiac arrhythmia in medical IoT

doi: 10.1038/s41598-026-43578-6

Figure Lengend Snippet: CLARITY-AI 2.0 overview and deployment context. (a) End-to-end hybrid architecture showing on-device ECG feature extraction, cloud inference, and the integrated explainability + security layers (SHAP-LLM explanations and intrusion detection) designed for security-aware edge cardiac monitoring. (b) Deployment scenario for continuous ECG streaming in a medical IoT setting, highlighting how predictions, explanations, and trust/security flags are delivered to enable interpretable and trustworthy decision support at the network edge.

Article Snippet: In contrast, recent advances in MIoT and wearable sensing technologies have enabled continuous, real-time electrocardiogram (ECG) monitoring, facilitating early disease detection and personalized health management , .

Techniques: Extraction

Journal: Scientific Reports

Article Title: Lightweight and interpretable edge intelligence AI with intrusion detection for trustworthy cardiac arrhythmia in medical IoT

doi: 10.1038/s41598-026-43578-6

Figure Lengend Snippet: Multi-source data segmentation and input representation. Visual overview of the data preparation pipeline and beat-level segmentation. The figure illustrates how 12-lead ECG signals are segmented and aligned into a single beat representation; the example shown is a beat from the PTB-XL dataset rendered consistently across all 12 leads for downstream feature extraction and modeling.

Article Snippet: In contrast, recent advances in MIoT and wearable sensing technologies have enabled continuous, real-time electrocardiogram (ECG) monitoring, facilitating early disease detection and personalized health management , .

Techniques: Extraction

Journal: Scientific Reports

Article Title: Lightweight and interpretable edge intelligence AI with intrusion detection for trustworthy cardiac arrhythmia in medical IoT

doi: 10.1038/s41598-026-43578-6

Figure Lengend Snippet: Local explanation case study (ANOMALY/PVC): SHAP + LLM report. Example of a model-specific explanation for an anomalous beat. (a) The ECG segment used for inference. (b) SHAP waterfall plot showing the dominant positive/negative feature contributions driving the anomaly decision. (c) The corresponding LLM-generated clinician-readable explanation produced from the SHAP evidence.

Article Snippet: In contrast, recent advances in MIoT and wearable sensing technologies have enabled continuous, real-time electrocardiogram (ECG) monitoring, facilitating early disease detection and personalized health management , .

Techniques: Generated, Produced

Journal: Scientific Reports

Article Title: Lightweight and interpretable edge intelligence AI with intrusion detection for trustworthy cardiac arrhythmia in medical IoT

doi: 10.1038/s41598-026-43578-6

Figure Lengend Snippet: Local explanation case study (NORMAL): SHAP + LLM report. Example explanation for a normal beat. (a) The ECG segment used for inference. (b) SHAP waterfall plot showing which features support the normal classification versus counter-evidence. (c) The final clinician-oriented LLM explanation grounded in the SHAP attribution list.

Article Snippet: In contrast, recent advances in MIoT and wearable sensing technologies have enabled continuous, real-time electrocardiogram (ECG) monitoring, facilitating early disease detection and personalized health management , .

Techniques:

Journal: Scientific Reports

Article Title: Lightweight and interpretable edge intelligence AI with intrusion detection for trustworthy cardiac arrhythmia in medical IoT

doi: 10.1038/s41598-026-43578-6

Figure Lengend Snippet: On-device efficiency on ESP32 (latency + footprint). On-device benchmark comparing CLARITY-AI 2.0 to a 1D-CNN baseline deployed on the same ESP32. The figure summarizes runtime feasibility and resource usage, showing that CLARITY-AI 2.0 is 11.7× faster and remains well below a 100 ms real-time constraint for beat-level inference, while also substantially reducing model/storage demands (energy results are detailed in Fig. ).

Article Snippet: In contrast, recent advances in MIoT and wearable sensing technologies have enabled continuous, real-time electrocardiogram (ECG) monitoring, facilitating early disease detection and personalized health management , .

Techniques:

Journal: medRxiv

Article Title: Pharmacological enhancement of glymphatic function in humans increases the clearance of Alzheimer’s disease-related proteins

doi: 10.64898/2026.03.10.26348048

Figure Lengend Snippet: The Dexmedetomidine Trial was conducted at Stanford University’s Clinical and Translational Research Unit (CTRU), while the ACX-02 Clinical Trial was conducted at both Stanford University’s CTRU and Washington State University’s Sleep and Performance Research Center. Both trials used a cross-over design - treatment followed by placebo - to assess the effect of DEX on the glymphatic clearance of Aβ and tau from the brain to the blood. (A) The Dexmedetomidine Trial enrolled nine participants, of whom eight completed both study visits. (B) The ACX-02 Trial enrolled 22 participants, eight at Stanford University and 14 at Washington State University. Of the eight participants at Stanford, six completed both study visits. At Washington State University, 11 of 14 participants completed both study visits. (C) Illustration of the determinants of glymphatic clearance of Aβ and tau during wake, NREM sleep, DEX and ACX-02 treatment. To capture these physiological determinants, participants were instrumented with ECG telemetry, percutaneous oxygen saturation (SpO₂), nasal cannula for low-flow oxygen and continuous end-tidal CO₂ (EtCO₂) monitoring, and a radial arterial line for continuous blood pressure monitoring and blood sampling (Philips IntelliVue MP50). An intravenous line was placed for drug or saline placebo infusion. Participants were also fitted with an investigational in-ear wearable device from Applied Cognition¹⁷ that measured key determinants of glymphatic function, including sleep features (hypnogram and spectral band power) by EEG, heart rate variability (HRV) by photoplethysmography (PPG), cerebrovascular pulse transit time (PTT cereb ) by impedance plethysmography (IPG), and brain parenchymal resistance (R P ) by dynamic electrical impedance spectroscopy.

Article Snippet: Participants were instrumented for monitoring by electrocardiography (ECG) telemetry, percutaneous oxygen saturation (SpO 2 ), nasal canula for administration of supplemental oxygen at 2 liters per minute (LPM) and continuous end-tidal CO 2 (EtCO 2 ) monitoring, and a 20 gauge (g) radial arterial catheter at the wrist for continuous systemic blood pressure measurement and blood sampling (Philips IntelliVue MP50 Patient Monitor).

Techniques: Clinical Proteomics, Sampling, Saline, Impedance Spectroscopy

Journal: Journal of Cardiovascular Magnetic Resonance

Article Title: Free-breathing three-dimensional whole-heart adiabatic T1ρ mapping for non-contrast tissue characterization at 0.55T

doi: 10.1016/j.jocmr.2025.102676

Figure Lengend Snippet: Framework of the proposed 3D adiabatic T1ρ mapping research sequence. (A) Pulse sequence diagram. T1ρ preparation is performed with adiabatic SL that consists of multiple pairs of HS pulses. Four interleaved 3D volumes are acquired in repeated cycles of four heartbeats with electrocardiogram triggered SPGR acquisition. The first three volumes (IMG1-IMG3) are prepared with saturation pulse with the same recovery time (T SAT ), while the last volume (IMG4) is prepared with saturation pulse without delay time. IMG2 and IMG3 are also prepared with adiabatic-SL consisting of 2 and 4 HS pulses with SL duration ( τ SL ) of 2 τ HS and 4 τ HS , respectively. 2D iNAVs are performed before data acquisition in each heartbeat, and a variable-density Cartesian trajectory with spiral-like profile order (VD-CASPR) is adopted with 4-fold under-sampling. (B) T1ρ map calculation. The four volumes with different contrast weighting (IMG1-IMG4) are used to calculate a 3D T1ρ map voxel-by-voxel using a 3-parameter fitting method. SL spin-lock, 2D two-dimensional, 3D three-dimensional, HS hyperbolic-secant, SPGR spoiled gradient echo

Article Snippet: ECG trigger signal was recorded from an external ECG monitoring system (Expression MR400, Philips Healthcare, Best, The Netherlands).

Techniques: Sequencing, Sampling