A modular, deep learning-based holistic intent sensing system tested with Parkinson's disease patients and controls

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Mobility-limiting conditions like Parkinson's disease impede task completion.
• Intent-sensing technology offers potential for assistive devices to aid users.
• Prior work includes Probabilistic Sensor Networks and time-segmented deep learning for intent prediction.

Purpose of the Study:

• To develop and test a novel multi-modal intent sensing algorithm.
• To combine Inertial Measurement Unit (IMU) and microphone sensor data.
• To interpret sensor data using time-segmented deep learning for activity classification.

Main Methods:

• Developed a novel algorithm for multi-modal intent sensing.
• Integrated IMU and microphone sensor data.
• Applied time-segmented deep learning for continuous intent prediction.
• Tested the algorithm on a dataset including non-disabled individuals and those with Parkinson's disease.

Main Results:

• Achieved 97.4% accuracy in determining user intent within 0.5 seconds of the initial intention.
• Intent determination accuracy improved monotonically to a maximum of 99.9918% during the activity.
• Successfully classified three activities of daily living.

Conclusions:

• Intent sensing is a viable technology for assistive medical devices.
• The proposed multi-modal approach enhances accuracy and speed of intent detection.
• This technology can significantly aid individuals with mobility-limiting conditions.