Hi!

Welcome again to AIMedily.

This week, I participated in an online session about AI in medicine.

What made it special wasn’t the topic (even though I love AI).

It was because my favorite middle school teacher invited me.

I still remember how creative his classes were.

The experience reminded me of the power of reconnecting.

AI is transforming everything around us - it can help us reconnect and be creative.

Now, it's time to check what’s in today's issue:

🤖AIBytes: Soft wearable robot to improve gait. Proprioceptive robot training in stroke. Exoskeleton + Brain computer interfaces.

🦾TechTool: BiomechGPT: Accurate movement analysis using only video.

🧬AIMedily Snaps: Don’t miss this week's links.

🧩TriviaRX: Find the new question (and the answer from last week).

🤖 AIBytes

Soft Wearable Robot (SWR) Enhances Gait in Stroke

This randomized crossover trial included 19 chronic stroke patients.

The intervention used a lightweight cable-driven soft wearable robot (SWR). Targeted hip flexion and extension.

🔬 Methods:

Researchers assessed each participant under three conditions:

No device (No SWR)
Wearing the device without power (Robot Off)
Wearing the device with active assistance (Robot On)

The robot employed real-time gait phase estimation and personalized assistive force. It was controlled via a mobile app.

Evaluations:

Gait performance: 10-Meter Walk Test (10MWT), 6-Minute Walk Test (6MWT)
Energy expenditure (oxygen cost and rate using COSMED K5)
3D kinematic and kinetic gait analysis
Gait deviation metrics

Each participant completed familiarization sessions before testing.

📊 Results:

Gait Performance:
- Walking speed increased by 0.11 m/s (Robot On vs. No SWR)
- 6MWT distance increased by 20.7 meters
- Both exceeded Minimal Clinically Important Differences

Energy Efficiency:
- Oxygen cost reduced by 14.9% in Robot On vs. No robot
- Metabolic savings: 0.57 vs. 0.67 mL/kg/m (p < 0.05)

Kinematics:
- Hip flexion at initial contact ↑ 39.4% (p = 0.002)
- Maximum hip flexion during swing ↑ 41.1% (p < 0.001)
- Knee flexion at initial contact ↑ 30.4% (p = 0.022)
- Pelvic hiking decreased by 80% (p = 0.001)

Spatiotemporal:
- Step length on the affected side increased (p = 0.005)
- Single support duration ↑ on affected side (p = 0.003)

Gait Deviation:
- No significant improvement in asymmetry

🔑 Key Takeaways:

A single session with the SWR led to clinically meaningful improvements in:

Walking speed and endurance.

Energy cost decreased significantly, making walking less metabolically demanding.
Kinematic gains in hip and knee angles reduced compensatory patterns. Gait asymmetry persisted, likely due to chronic compensatory motor strategies.
The lightweight SWR design enhanced comfort and minimized the added metabolic burden.

🔗Han SH, Choi S, Ko C, et al. Efficacy of a Soft Wearable Robot for Hip Assistance in Chronic Stroke Patients: A Randomized Crossover Trial. IEEE Trans Neural Syst Rehabil Eng. Published online June 13, 2025. doi:10.1109/TNSRE.2025.3577600

Self-Guided Proprioceptive Robot Training Boosts Motor Recovery

🔬 Methods:

This study involved:

5 individuals with chronic stroke
5 age-matched controls

They received a single 38-minute session of self-guided proprioceptive training using:

KINARM exoskeleton paired with a joystick.

Participants used their less-affected arm to guide passive movement of their more-affected limb.
The task required reliance on real-time proprioceptive feedback with no visual feedback during the movement.

Pre/post robotic assessments:

Visually Guided Reaching (VGR) – motor control
Arm Position Matching (APM) – proprioception

Clinical assessments for stroke participants included:

📊 Results:

Motor Control (VGR):
- Stroke group improved significantly ( p < 0.001)
- No significant change in controls (p = 0.87)
- Stroke group improved in 5 motor parameters:
  - Posture speed
  - Reaction time
  - Initial direction angle
  - Min–max speed difference
  - Movement time
Proprioceptive Accuracy (APM):
No significant change in the overall APM Task Score.
But stroke participants showed meaningful improvements in specific spatial alignment measures.

🔑 Key Takeaways:

Motor function improved significantly after one proprioceptive training session in all individuals with stroke.
Targeted proprioceptive gains (spatial alignment and endpoint control) suggest sensory retraining—even when the APM score did not change significantly.

🔗Tulimieri DT, Kim G, Hoh JE, Sergi F, Semrau JA. A pilot study for self‑guided, active robotic training of proprioception of the upper limb in chronic stroke. J NeuroEngineering Rehabil. 2025;22(1):130. doi: 10.1186/s12984-025-01660-6

Exoskeleton-Guided Passive Movement Elicits Standardized EEG Patterns for BCIs (Brain Computer Interface) in Stroke

Researchers recorded and analyzed EEG signals during exoskeleton-assisted movements and voluntary movements.

🔬 Methods:

Participants: 20 healthy adults and 10 stroke survivors.

Procedure:

Researchers recorded EEG signals during voluntary and passive hand movements.
The robotic exoskeleton guided the passive movement.
They completed 80 trials (40 per hand). Following a structured protocol: 0.5 s cue → 2 s movement → 3 s rest.

They tested 2 AI Models:

CSP (traditional model)
EEGNet (deep learning model)

📊 Results:

In healthy subjects, EEG signals were stronger and more reliable during Passive Movement.
Stroke patients showed normalized EEG during passive movement but not during active
Voluntary movement in stroke patients showed disrupted mu/beta rhythms.
The system could detect hand movement intentions from brain signals.
The same model was accurate when trained on other people’s data:
86% (healthy) / 72.6% (stroke) —without calibration.

🔑 Key Takeaways:

Passive exoskeleton-guided movements generate stronger and more consistent EEG responses than voluntary movements. Even in stroke patients with severe motor impairments.
BCI model (EEGNet) distinguishes hand movement intentions—identifying affected vs unaffected hands. 84.2% accuracy in stroke patients.
Deep learning models trained on passive EEG data generalize well.

🔗 Zhang X, Xie L, Liu W, et al. Exoskeleton-guided passive movement elicits standardized EEG patterns for generalizable BCIs in stroke rehabilitation. J Neuroeng Rehabil. 2025;22(1):97. doi: 10.1186/s12984-025-01627-7

🦾TechTool

BiomechGPT: Accurate Movement Analysis Using Only Video

🔬 Methods:

BiomechGPT, is a multimodal AI model designed to estimate human biomechanics from video alone—without sensors or markers.

The researchers trained the model after 30 hours of biomechanics data from nearly 500 participants. Many of them had movement impairments.

With the data collected, they tokenize movement trajectories (convert 3D human movement motion into discrete tokens that can be processed by the LLM).

📊 Results:

Achieved ~90% accuracy in predicting key biomechanical metrics, including joint kinematics and dynamics.
Generalized across varied movement tasks and participant populations, including healthy and impaired individuals.
Outperformed traditional computer vision models by 15% in low-data scenarios. Showing strong data efficiency.

🔑 Key Takeaways:

Markerless and scalable: Provides accurate motion analysis from video. No wearables or lab equipment needed.
Demonstrate high performance on activity recognition, identifying movement impairments, diagnosing and scoring clinical outcomes, and measuring walking.
Efficient and adaptable: Works with limited training data, making it suitable for rehabilitation settings with small datasets.

🔗 Yang R, Kennedy A, Cotton RJ. BiomechGPT: A Multimodal Foundation Model for Human Biomechanics from Video. arXiv preprint. 2025. doi: 10.48550/arXiv.2505.18465

🧬AIMedily Snaps

🧩TriviaRX

What’s the oldest known prosthetic limb ever discovered?

A) A Roman iron hand

B) An Egyptian wooden toe

C) A bronze Greek arm

D) A bamboo leg from China

The answer will be in the next issue.

Now, the answer to last week's question…drumrolls 🥁:

✅C) Placing live torpedo fish on the affected body part.

In approximately 46 AD, the Roman physician Scribonius Largus documented using live electric rays—often referred to as “torpedo fish”—for therapeutic electrical stimulation to relieve pain from gout and headaches.

We’re done for today.

If you enjoy AIMedily, maybe your colleagues will enjoy it too!

Please help me share it 😊: