Sponsored by
Looking for unbiased, fact-based news? Join 1440 today.
Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.
Hi!
Welcome back to AIMedily!
This weekend I spent a few days in New York City, where my dad did his surgery residency in the 70s.
I passed by the hospital he worked at.
The same building from the outside. But now, on the inside, surgeries are different because of technology.
AI and tech keep changing how we practice medicine.
Technology has helped us to improve diagnosis, quality, accuracy, and functional outcomes.
But humans will always be at the center of medical attention. From patients to healthcare workers.
Now, time to check today's issue.
🤖AIBytes:
Neuroplasticity with robot training in cerebral palsy.
A systematic review on Robot-Assisted Training.
Machine Learning and Depth Cameras in Physiotherapy
LLMs for board exams in PM&R?
🦾TechTool: NeuroFlex: a soft rehab glove + EEG.
🧬AI Medily Snaps: Check out a non-profit that supports academics. How AI is disrupting medicine, a neural prosthesis seminar, and more.
🧩TriviaRX: Find the new question (and the answer from last week).
🤖 AIBytes
🔬 Methods:
Participants: 30 children (ages 4–6) with spastic diplegic cerebral palsy. Classified as GMFCS levels II–III.
Protocol: Each child completed three conditions while EEG data were recorded:
Rest (REST) – no specific task.
Motor Imagery (MI) – imagining moving their legs without actual movement.
Motor Imagery with Brain–Computer Interface and Robotic Training (MI‑BCI). Imagining movement while controlling a robotic lower-limb device through brain signals.
EEG Analysis:
Researchers identified four EEG microstates (A–D), which represent different functional brain networks.
Microstate A: Rest
Microstate B: Cognitive readiness
Microstate C: Default-mode (internal thoughts)
Microstate D: Motor attention and planning
They measured how often and how long these microstates appeared. And how the brain transitioned between them.
Functional connectivity (FC) was also analyzed to understand how different brain regions coordinate.
📊 Results:
In MI and MI‑BCI conditions :
Increased duration and coverage of microstate D (associated with motor attention).
Decreased microstate A (indicating less rest of the brain) compared to REST (P < 0.05).
The brain's transitions shifted from resting to motor-focused during MI and MI‑BCI.
Compared to Motor Imagery alone, MI‑BCI training:
Reduced default-mode or passive brain activity.
Increased frequency of mental readiness for movement.
Functional Connectivity:
MI‑BCI provides wider, stronger cortical network engagement.
Especially in attention, motor planning, and execution.
🔑 Key Takeaways:
MI‑BCI training increases brain activity linked to motor planning.
MI-BCI training shows stronger connectivity. The brain function was more synchronized and engaged.
These changes suggest neuroplasticity effects with MI-BCI training.
EEG microstates and connectivity patterns could serve as brain biomarkers to track progress for children with CP.
🔗Qi W, Zhang Y, Su Y, et al. Exploring cortical excitability in children with cerebral palsy through lower limb robot training based on MI‑BCI. Sci Rep. 2025;15:12285. doi:10.1038/s41598-025-96946-z
🔬 Methods:
Design: Systematic review and meta-analysis of 12 randomized controlled trials.
Participants: 651 adult stroke survivors.
Intervention: Robot-assisted lower-limb rehabilitation vs. standard therapy.
Tools:
Robotic devices included exoskeletons and end-effector robots.
Evaluations:
Analysis: Meta-analysis using RevMan 5.4. Follows PRISMA guidelines.
📊 Results:
Motor Function (FMA-LE):
Robot training improved motor scores by +3.17 points ( p < 0.00001).
End-effector robots: +6.42 points
Exoskeletons: +2.41 points
Walking Endurance (6MWT):
Participants walked 13.32 meters more ( p = 0.0007).
Exoskeletons: +19.52 meters
End-effectors: Not significant
Balance (BBS):
Improved by +6.98 points overall (p = 0.0005).
Exoskeletons: +9.31 points
End-effectors: +2.97 points
Gait Coordination (TUGT):
No significant improvement (p = 0.59).
🔑 Key Takeaways:
Robot-assisted rehab improves leg motor function. End-effector devices showing the greatest gains.
Exoskeletons significantly boost walking distance and balance more than end-effectors.
No meaningful improvement in gait speed or coordination. Robot training may need to be combined with other modalities.
Robotic devices offer a measurable clinical benefit in stroke recovery. Especially for balance and walking endurance.
🔗Wang H, Shen H, Han Y, Zhou W, Wang J. Effect of robot-assisted training for lower limb rehabilitation on lower limb function in stroke patients: a systematic review and meta-analysis. Front Hum Neurosci. 2025;19:1549379. doi:10.3389/fnhum.2025.1549379
🔬 Methods:
This systematic review (2020–2024) evaluated 371 publications across four databases (Web of Science, Scopus, PubMed, ADS).
Researchers selected 18 studies focused on depth camera–based physiotherapy assessments.
📊 Results:
Depth camera types: Kinect devices (65.4%), RealSense, structured-light, and time-of-flight.
Data streams: RGB-D imagery (55.6%) and skeletal joint data (27.8%).
Traditional Machine Learning approaches (44.4%)
Deep Learning Models (41.7%)
Settings:
50% lab-based
33.4% clinical
22.3% remote/home assessments.
🔑 Key Takeaways:
Depth cameras, especially Kinect, are the primary tool for non-contact physiotherapy movement monitoring.
Machine Learning applications include classic algorithms and deep models. Both are useful for assessing rehabilitation movements.
To translate lab success into clinical utility, future studies must include:
Larger clinically diverse cohorts. Stronger clinical testing. Strandarized datasets.
🔗 Zhou Y, Rashid F’AN, Mat Daud M, Hasan MK, Chen W. Machine Learning-Based Computer Vision for Depth Camera-Based Physiotherapy Movement Assessment: A Systematic Review. Sensors. 2025;25(5):1586. doi: 10.3390/s25051586
🔬 Methods:
GPT-3.5 and GPT-4o on a set of 744 board-style PM&R questions.
They covered key domains (e.g., stroke, SCI, pediatrics, pain, pharmacology).
Each model answered the entire question bank three times to assess:
Accuracy
Consistency
📊 Results:
GPT-3.5 performance:
Run 1: 56.3% correct
Run 2: 56.5%
Run 3: 56.9%
GPT-4o performance:
Run 1: 83.6%
Run 2: 84.0%
Run 3: 84.1%
GPT-4o significantly outpaced GPT-3.5 across all PM&R knowledge subdomains. Demonstrating both higher accuracy and stable consistency.
🔑 Key Takeaways:
GPT‑4o achieves ≥ 83% accuracy on PM&R questions. 27 percentage points better than GPT‑3.5.
The consistency confirms GPT‑4o’s reliability.
Superior performance across subspecialties (e.g., stroke, rehab, pharmacology) shows broad clinical knowledge.
These results suggest LLMs like GPT‑4o could support board review, training, and education. Clinical caution remains essential.
🔗 Bitterman J, D'Angelo A, Holachek A, Eubanks JE. Advancements in large language model accuracy for answering physical medicine and rehabilitation board review questions. PM R. Published May 2, 2025. doi:10.1002/pmrj.13386
🦾TechTool
🔬 Methods:
This feasibility study introduced NeuroFlex.
A soft robotic glove designed to support hand rehabilitation.
The glove translates EEG-based motor imagery (MI) into real-time movement.
Participants: 3 healthy adults
Protocol: Each participant completed 18 trials under three conditions:
Motor execution (actual fist movement)
Motor imagery (MI) – imagining the same movement
Rest
Each condition lasted 16 sec.
Participants wore a 16-channel dry-contact EEG headset.
AI Model and Glove Control:
A transformer-based deep learning model was trained to detect motor imagery (MI) vs. rest.
When MI was detected, a microcontroller activated the glove’s pneumatic actuators, producing a fist movement—controlled by brain signals.
📊 Results:
Classification Accuracy: 85.3% for detecting Motor Imagery.
AUC (Area Under Curve): 0.88 — indicating strong distinction between Motor Imagery and rest.
🔑 Key Takeaways:
NeuroFlex successfully translates imagined hand movements into glove motion.
Uses Non-invasive EEG and AI.
Neuroflex is a tool that can be used for future clinical applications on impaired hand function.
🔗Zare S, Beaber SI, Sun Y. NeuroFlex: Feasibility of EEG-Based Motor Imagery Control of a Soft Glove for Hand Rehabilitation. Sensors. 2025;25(3):610. doi:10.3390/s25030610
🧬AIMedily Links
Laude. A non-profit that supports academics who want to turn their research into products.
A Neural Prosthesis seminar series from the Cleveland Health Center.
JMIR Rehabilitation and Assistive Technologies Receives Inaugural Journal Impact Factor (check the most cited articles inside)
🧩TriviaRX
What historical medical treatment for paresis involved infecting patients with malaria (won a Nobel Prize)?
a) Hydrotherapy
b) Insulin shock therapy
c) Pyrotherapy
d) Lobotomy
✅ Last week correct Answer: B) Wooden toe (Egypt, 950–710 BCE)
Found on a female mummy, it’s considered the earliest functional prosthetic known.
[Ref: Journal of Prosthetics and Orthotics, 2000.]
Did you guess the correct answer?
We’re done.
One huge favor, please share AIMedily:
You can forward this email ⏩, or
Click on this link AIMedily and share as a message from your phone 📲
With your help we can become a larger and stronger community 🥰.
If you have articles on AI and Rehabilitation and want to share them, reply to this email 📤.
That’s all for today. See you next week.
Itzel Fer, MD PM&R
Forwarded this email? Sign up here
How did you like today's newsletter?
Start learning AI in 2025
Keeping up with AI is hard – we get it!
That’s why over 1M professionals read Superhuman AI to stay ahead.
Get daily AI news, tools, and tutorials
Learn new AI skills you can use at work in 3 mins a day
Become 10X more productive
