Discriminating Free Hand Movements Using Support Vector Machine and Recurrent Neural Network Algorithms.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Computer Science

Background:

• Accurate decoding of natural hand movements is crucial for human-computer interaction.
• Challenges exist in measuring complex hand movements and decoding dynamic data.
• Potential applications include diagnosing motor diseases and monitoring rehabilitation.

Purpose of the Study:

• To develop and compare two algorithms for discriminating subtle differences in hand movement sequences.
• To evaluate the performance of Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) neural networks for hand movement decoding.

Main Methods:

• Recorded hand movement data from 34 adults (17 younger, 17 older) using an exoskeletal data glove.
• Participants performed six distinct hand movement tasks.
• Developed and applied SVM with dynamic time warping and LSTM algorithms for classification.

Main Results:

• Both SVM and LSTM achieved high accuracy in across-subject classification of hand movements.
• The SVM-based approach significantly outperformed LSTM in within-subject classification.
• SVM demonstrated robust performance, particularly for generalizing across age groups.

Conclusions:

• The SVM-based algorithm is a promising tool for decoding hand movements, especially for applications requiring cross-age generalizability.
• This approach can aid in detecting motor disorders and tracking their progression.
• Further development could enhance human-computer interaction and clinical diagnostic tools.