Smartphone-based biofeedback system for upper-limb motor rehabilitation using low-cost motion sensors and machine learning.

Project Abstract

The proposed study develops a smartphone-based biofeedback system to enhance upper-limb motor rehabilitation by leveraging low-cost inertial measurement units (IMUs) and machine learning-driven personalization. The core objective is to deliver real-time, quantitative feedback on motor performance to stroke survivors and patients with motor impairments, enabling remote, engaging, and data-driven therapy. The system integrates readily available smartphone sensors (accelerometer, gyroscope, and magnetometer) with a compact, add-on low-cost motion-sensing module to capture kinematic data during therapeutic exercises. A hybrid data pipeline combines edge processing on-device with cloud-based services to ensure low latency, privacy-preserving data transmission, and scalable analytics. Advanced signal processing techniques extract clinically meaningful features such as trajectory smoothness, angular velocity, jerk, movement parity, and adaptive range-of-motion metrics. These features feed a personalized machine learning model that adapts exercise difficulty, provides corrective cues, and generates motivation through gamified feedback. The study investigates supervised and semi-supervised learning approaches to classify movement quality, detect compensatory strategies, and predict rehabilitation progress, with attention to class imbalance and inter-subject variability. A randomized, crossover trial will evaluate the system’s efficacy against conventional home-based therapy, measuring outcomes including Fugl-Meyer Assessment for the upper extremity, Box and Blocks test, grip strength, and participation-level metrics. Secondary outcomes will assess adherence, user satisfaction, perceived exertion, cognitive load, and usability (System Usability Scale and Net Promoter Score). The research also examines the feasibility of scalable deployment in low-resource settings, addressing device heterogeneity, offline operation, data privacy, and cost-effectiveness. An embedded validation framework will compare automated assessments against clinician-rated scores, establishing reliability and validity in real-world environments. The expected contributions include (i) a modular, interoperable biofeedback system that leverages ubiquitous smartphone hardware to democratize access to rehabilitation, (ii) a machine learning pipeline capable of adapting to individual impairment profiles and exercise regimens, and (iii) evidence on the clinical utility of remote, sensor-driven rehabilitation with real-time feedback. The study will also explore normative baselines for movement metrics across different etiologies and stages of recovery to inform personalization strategies. Potential limitations involve variability in sensor placement, user adherence, and network connectivity, which will be mitigated through robust calibration protocols, offline-first design, and adaptive feedback that emphasizes intrinsic motivation. Overall, the project aims to empower patients with timely, objective, and actionable information to optimize upper-limb rehabilitation outcomes while reducing the burden on healthcare systems.

Project Overview

What This Project Is About

---

A straightforward exploration of using a smartphone to support rehabilitation of the arm. The project tests whether affordable motion sensors in a phone can track arm movements and provide helpful feedback to guide exercises, with simple machine learning to tailor feedback to the user.

---

---

The Problem It Addresses

---

Many rehabilitation options are costly or require frequent clinic visits. Patients often struggle to stay motivated or perform exercises correctly at home. This project seeks a low-cost, accessible system that helps people practice movements safely and effectively outside the clinic.

---

---

Objectives of the Project

---

1. Assess whether a smartphone can accurately measure upper-limb movements using its built?in sensors.
2. Develop a simple feedback mechanism to guide correct exercise form.
3. Implement a lightweight machine learning model to adapt feedback to the user over time.
4. Evaluate user ease of use and perceived usefulness through a small user study.
5. Propose practical guidelines for deploying the system in home rehabilitation.

---

---

What You Will Do Step by Step

---

1. Review existing rehab methods and sensor basics. 2. Collect movement data from participants performing target exercises using a smartphone. 3. Label data with correct vs. incorrect form. 4. Build a simple feature set from sensor readings. 5. Train a lightweight model to detect form errors. 6. Create real-time feedback prompts in the app. 7. Test usability with a user group. 8. Analyze accuracy and user satisfaction.

---

---

Expected Outcome

The project should deliver a prototype smartphone app that can monitor upper-limb exercises, provide clear feedback to improve form, and show preliminary evidence that the approach is usable and beneficial for home rehab.