Combined methods for neuroprosthetics improvement

• Multi-method integration may improve neuroprosthetics performance or robustness.¹
• AIP reports suggest a physics/engineering angle for combining methods in neuroprosthetic design.^{1 1}

Weekly enrichment (2026-07-20)

• The underlying source is a Perspective by Leong et al., “Optimization frameworks for bespoke sensory encoding in neuroprosthetics,” in APL Bioengineering, highlighted by an AIP SciLight; senior author Solaiman Shokur.^{2 3}
• The focus is somatosensory restoration: artificial sensors capture environmental information that is encoded into stimulation delivered through neural interfaces, restoring sensation to limbs after spinal cord injury or stroke.²
• Three optimization frameworks are defined by the response × optimizer pairing: explicit (perceptual data plus algorithm), physiological (physiological signals plus algorithm), and self-optimized (perceptual data with the subject as the optimizer).^{3 4}
• Explicit frameworks (psychophysical tests or questionnaires) are the gold standard but slow, whereas physiological frameworks iterate rapidly via computer algorithms but require a good proxy measure that correlates with perception.^{2 3}
• The proposed combined pipeline uses the physiological framework for rapid coarse tuning to narrow the parameter space, then an explicit or self-optimized framework for fine tuning, balancing fast convergence with perceptual accuracy.^{3 4}
• The authors suggest EEG features that correlate with sensations across subjects could serve as robust proxy measures in future physiological optimization.³
• A complementary line of work combines biophysical models with machine-learning surrogates to co-optimize intraneural (TIME) implant geometry and multipolar stimulation protocols, yielding substantial in-silico selectivity gains.⁵
• On the motor side, hybrid EMG-EEG fusion improves robustness: a fatigue-adaptive Bayesian-fusion elbow rehab robot reached 94.5% accuracy (vs 88.5% for EMG-only) with under 500 ms latency, and an MI-based EEG-EMG connectivity network reached 94.33% (vs 73.89% EEG and 89.16% EMG alone).^{6 7}

Footnotes

1. https://news.google.com/rss/articles/CBMigwFBVV95cUxNcjZJMWFSMkVQZUJqRXBhdUIwc2xZWFdZeUNza01kS3ktS05QTXJxUXFPOHBLbERIdkZqRHprbHdyU1RTNnFfRTY0YWZTdmgwUGhFbWFiODVKaXBvdXBEU2dyUU5wSnNSUkdGNHdyZ1BDV043RGliSE4zbThyYUVxOWdqWQ?oc=5 ↩ ↩² ↩³

2. https://www.aip.org/scilights/combined-methods-could-improve-neuroprosthetics ↩ ↩² ↩³

3. https://doi.org/10.1063/5.0249434 ↩ ↩² ↩³ ↩⁴ ↩⁵

4. https://pmc.ncbi.nlm.nih.gov/articles/PMC12094799/ ↩ ↩²

5. https://doi.org/10.1088/1741-2552/ace219 ↩

6. https://doi.org/10.1038/s41598-025-24831-w ↩

7. https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2025.1532099/full ↩