Hi!
Welcome again to AIMedily.
Lately, there’s controversy when comparing AI vs physicians on diagnosis (check the article below ⬇️).
But medical attention is not a diagnosis.
Medical care encompasses clinical reasoning, physical examination, treatment planning, analysis, etc.
Yet at the center, there’s one essential element: human compassionate care.
AI is here to facilitate and improve medical diagnosis. But it can’t replace human connection.
Now, let’s dive into today’s issue.
.🤖AIBytes: An AI wearable navigation system for blind people, AI vs physicians, who does better?, and a robotic glove for upper limb rehab.
🦾TechTool: Markerless motion capture in upper limb post-stroke.
🧬AI Medily Snaps: Personalized medicine with AI. The use of AI in clinical trials. A global perspective on AI in healthcare and the next conferences.
🧩TriviaRX: Did you get the right answer? Let’s check it out.
🤖 AIBytes
🔬 Methods
Wearable navigation system for blind or visually impaired (BVI) people.
Study Design: Pilot experimental
Participants: 10 blind or low-vision individuals (6 males, 4 females; mean age 56 ± 11 years).
Had experience using mobility aids (canes or guide dogs).
A wearable helmet with an IMU (inertial measurement unit), cameras, and an ultrasonic radar + mobile phone.
AI navigation assistant (Deep Learning).
Tactile and auditory feeback (vibration)
Virtual companionship.
Participants walked through structured indoor and outdoor environments with and without the MNVC system.
📊 Results
Obstacle avoidance:
Participants had 32% fewer collisions while using the MNVC system.
Walking efficiency:
Mean walking time was 18.3% faster with the device (p < 0.05).
User satisfaction: High usability
🔑 Key Takeaways
The MNVC system significantly reduced obstacle collisions and speed.
Intuitive device, requiring minimal training.
Feedback improved orientation and confidence.
Provides real-time, safe and reliable navigation for blind people.
Has potential to improve autonomy and safety.
🔗 XU J, Wang C, Li Y, Huang X, Zhao M, Shen Z, Liu Y, Wan Y, Sun F, Zhang J, et al. Multimodal Navigation and Virtual Companion System: A Wearable Device Assisting Blind People in Independent Travel. Sensors 2025: 25(13):4223. Doi: 10.3390/s25134223
🔬 Methods
Researchers created:
SDBench, an interactive framework for evaluating AI agents and physicians.
Simulates real-world diagnostic reasoning using 304 complex clinical cases from NEJM (2017–2025).
MAI-DxO, an AI "physician team" acting as a diagnostic orchestrator.
With SDBench and MAI-DxO, they tested:
Physicians: 21 U.S. and U.K. generalists (median 12 years’ experience); no specialists included. Each one completed 36 cases.
MAI-DxO + ChatGPT-3.5
Process:
Physicians and AI agendas received a short case abstract. Then they could:
Request additonal details
Order tests
Until they could provide a diagnosis.
Costs calculated using U.S.-based pricing data.
The test set consisted of 56 NEJM cases from 2024–2025.
Physicians were not allowed to use the internet, electronic records, language models, or clinical references—unlike typical real-world practice.
(To prevent answer leakage from online NEJM cases access)
📊 Results
Physicians: Diagnostic accuracy: 19.9%, Cost per case: $2,963
GPT-4o: Accuracy: 49.3%, Cost: $2,745
GPT-o3: Accuracy: 78.6%, Cost: $7,850
MAI-DxO+ChatGPT-03 (budget-optimized): Accuracy: 79.9%, Cost: $2,396
MAI-DxO+ChatGPT-03: Accuracy: 81.9%, Cost: $4,735
MAI-DxO improved accuracy and cost across all LLMs tested (p < 0.005) for most configurations.
🔑Key Takeaways
AI-led systems outperformed generalist physicians 4x in diagnostic accuracy on difficult NEJM CPC cases.
Physicians lacked real-world tools like internet access or specialist referrals, which limited their performance.
MAI-DxO reduced over-testing and cost.
Potential tool to diagnose in solo or resource-limited settings (lack of access to specialists or second opinions)
🔗Nori H, Daswani M, Kelly C, et al. Sequential Diagnosis with Language Models. arXiv. 2025;2506.22405v2.https://doi.org/10.48550/arXiv.2506.22405
Evaluating the effectiveness of the Gloreha robotic glove for upper-extremity rehabilitation after stroke (Function and Activities of Daily Living).
🔬 Methods
Study Design: Systematic review of randomized controlled trials (RCTs).
Participants:
Adult stroke patients (83 patients from 3 RCTs)
Acute/subacute to chronic stages.
Interventions:
2 studies: Gloreha glove + conventional therapy(CT) vs CT.
The glove was used in passive movement (finger counting, making a fist, pinching and finger movements).
1 study: Gloreha glove vs CT(Gloreha in active-assisted and game mode)
Treatment for both interventions: 30-60 min per session, 2-5/week, 3-6 weeks.
Assesments:
Fugl-Meyer Assessment (FMA-UE)
Motricity Index (MI)
Grip/pinch strength
Box and Block Test (BBT)
9-Hole Peg Test (NHPT)
Quick-DASH
Barthel Index (BI)
Visual Analog Scale (VAS)
📊 Results
Motor Function:
Gloreha groups significantly improve:
Motricity Index (p = .002)
Fugl-Meyer-UE: proximal domain (p = .03)
Dexterity & Strength:
9-hole peg test in Gloreha groups improved significantly over control (p=.009)
Box and block test; no differences between groups
Grip strength mixed results
Function and ADLs:
Quick-DASH improved significantly in Gloreha groups vs controls (p = .048).
Barthel Index im[proved significant within-group improvements, but inconsistent between groups,
🔑 Key Takeaways
Gloreha significantly improves upper limb motor function after stroke. Especially in the early recovery stages.
Dexterity and strength gains, particularly in fine motor control.
Improvement in Activities of Daily Living (ADL) was less consistent.
Future research should explore task-oriented treatment and long-term effects on ADL.
🔗 Thawisuk C, Apichai S, Chingchit W, et al. Effectiveness of robot-assisted upper extremity function training (Gloreha) on upper extremities function after stroke: systematic review. JMIR Rehabil Assist Technol. 2025;12(1):e68268. doi:10.2196/68268
🦾TechTool
🔬 Methods
Study Design: Cross-sectional experimental study comparing:
Upper-limb Markerless Motion Capture in stroke survivors and healthy controls across indoor and outdoor settings.
Participants:
50 individuals with chronic stroke (mean age: 58.9 ± 11.7 years)
49 age-matched healthy controls (mean age: 60.2 ± 8.5 years)
Tools:
A custom markerless motion capture (MMC) system.
Built with an iPad Pro + LiDAR-remote sensing method (used in topography).
Used ARKit6 (motion tracking for augmented reality) and RealityKit (3D simulation).
They tracked movements via a convolutional neural network (CNN) algorithm.
Assessments:
7 standardized upper-limb tasks using affected and unaffected limbs.
Fugl-Meyer Assessment for Upper Extremity (FMA-UE)
Wolf Motor Function Test (WMFT)
Functional Test for the Hemiplegic Upper Extremity (FTHUE)
📊 Results
Kinematic features were highly correlated with scores from clinical motor assessments.
All machine learning models achieved ≥ 85% sensitivity for classifying motor function.
🔑 Key Takeaways
Markerless motion capture reliably detects: Upper-limb deficits and asymmetries in stroke survivors. Indoors and outdoors.
Strong correlation with standard clinical tests.
Machine learning helps classify function accurately.
Portable MMC systems offer real-time remote monitoring.
🔗Lam WWT, Fong KNK, Chien CW. Upper limb kinematic measurement using markerless motion capturing (MMC) in stroke survivors: A cross-sectional experimental study. Digit Health. 2025;11:20552076251342009. doi:10.1177/20552076251342009
🧬AIMedily Snaps
🧩TriviaRX
Which 20th-century event accelerated the development of prosthetics and physical rehab as a specialty?
A) Spanish Flu
B) World War I
C) World War II
D) Polio Epidemic
✅The answer will be in the next issue!
👉Now, the answer from last Trivia:
B) EEG-based cursor control.
The earliest human BCI experiment used EEG signals to move a cursor on a screen, demonstrated by Dr. Jacques Vidal in 1973
Did you get it right?
That’s all for today!
Please send me feedback, what you like (or dislike) about the newsletter. This helps me create better content for you. ➡️Reply to this email.
Thank you!
See you next week,
Itzel Fer
MD PM&R
P.S. You can forward this email to your colleagues or share this link on your phone📱.
Forwarded this email? Sign up here
