A Wideband Multimodal Flexible Sensor Integrating Vertical Graphene and Sea Urchin-Like Nanoparticles for Post-Stroke Rehabilitation

 Do you really think your competent? doctor has enough functioning brain cells to get this into your stroke hospital?

Do you prefer your doctor and hospital incompetence NOT KNOWING? OR NOT DOING?

A Wideband Multimodal Flexible Sensor Integrating Vertical Graphene and Sea Urchin-Like Nanoparticles for Post-Stroke Rehabilitation

Geng Zhong Qingzhou Liu, Yunjun Huang, Haoyang Geng, Tailin Xu, First published: 21 August 2025 https://doi.org/10.1002/adma.202508206 

Stroke is a leading cause of long-term disability worldwide, with post-stroke aphasia significantly impairing communication and social interaction. Traditional rehabilitation devices are often bulky, expensive, and impractical for daily use, particularly in speech recovery, where accessible and effective solutions remain limited. To address this challenge, this study introduces a portable and wearable sensor system for stroke-induced aphasia rehabilitation. The proposed sensor integrates a flexible, ultrasensitive, and durable dual-sensor system comprising an Ag-MnO2-based sea-urchin-like nanoparticle pressure sensor to detect high-frequency vocal vibrations and a vertical graphene/polydimethylsiloxane (VGr/PDMS) strain sensor to capture low-frequency muscular movements. The sensors, integrated into a flexible circuit, employ an encoder-cycle-consistent generative adversarial networks (CycleGAN) model that recognizes users' intent and recovers voice, significantly reducing dependency on large-scale labelled datasets. Experimental results demonstrate accurate intent recognition with accuracies for certain commands exceeding 95%. The reconstructed speech exhibits improved naturalness based on objective and perceptual evaluations, highlighting potential clinical utility in enhancing daily communication and interaction for stroke survivors.

Graphical Abstract

This work presents a speech reconstruction framework based on a dual-sensor system that collects wideband signals from vocal vibrations and throat muscle movements. An encoder-cycle-consistent generative adversarial networks (CycleGAN) model maps the signals to speech without requiring a large-scale paired training dataset. The system achieves over 95% intent recognition accuracy and generates natural-sounding speech with a mean opinion score of 2.9.

Conflict of Interest

The authors declare no conflict of interest.