Scholarlink

Virtual reality (VR) can elicit controlled motor behaviours and capture rich hand-motion signals for digital biomarker research in Parkinson’s disease (PD). However, these recordings are clinically sensitive, motivating privacy-preserving analytics. This paper presents a prototype that combines (i) VR assessment minigames for PD-oriented motor elicitation and logging, (ii) a prototype using a fully connected neural classifier trained on standardized hand-motion features extracted from said VR software, and (iii) an adaptation for privacy-preserving inference via standard Secure Multi-Party Computation (MPC) protocols like Garbled Circuits (GCs) using fixed-point arithmetic. We demonstrate the validity of this setup for hand motion data collection, meaningful analysis inference and classification. Additionally, we provide a path forward for encrypted inference that preserves input/model confidentiality while maintaining functional classification and a practical path toward secure, reproducible VR-based motor assessment in PD.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

VR-Based Hand-Motion Assessment for Parkinson’s Disease: A Prototype Adapted for Secure Privacy-Preserving Neural Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nikola Hristov-Kalamov,
  • Laura García-Martín,
  • Raúl Fernández-Ruiz,
  • Bozhidar Petrov-Valchev,
  • Victoria Mora-de la Torre,
  • Agustín Álvarez-Marquina,
  • Rafael Martínez-Olla,
  • Daniel Palacios-Alonso

摘要

Virtual reality (VR) can elicit controlled motor behaviours and capture rich hand-motion signals for digital biomarker research in Parkinson’s disease (PD). However, these recordings are clinically sensitive, motivating privacy-preserving analytics. This paper presents a prototype that combines (i) VR assessment minigames for PD-oriented motor elicitation and logging, (ii) a prototype using a fully connected neural classifier trained on standardized hand-motion features extracted from said VR software, and (iii) an adaptation for privacy-preserving inference via standard Secure Multi-Party Computation (MPC) protocols like Garbled Circuits (GCs) using fixed-point arithmetic. We demonstrate the validity of this setup for hand motion data collection, meaningful analysis inference and classification. Additionally, we provide a path forward for encrypted inference that preserves input/model confidentiality while maintaining functional classification and a practical path toward secure, reproducible VR-based motor assessment in PD.