Handwritten character classification from EEG through continuous kinematic decoding.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Restoring communication for individuals with limited mobility is a critical challenge.
• Handwriting classification from neural signals offers a potential communication pathway.

Purpose of the Study:

• To classify ten handwritten letters (a,d,e,f,j,n,o,s,t,v) using non-invasive neural signals (EEG).
• To investigate the neural correlates of handwriting and compare classification methods.
• To explore a novel two-step approach combining continuous movement decoding and subsequent classification.

Main Methods:

• Direct classification of letters from low-frequency and broadband electroencephalogram (EEG).
• A two-step approach: continuous decoding of hand kinematics followed by letter classification.
• Analysis of neural signal components and hand movement kinematics.

Main Results:

• The two-step approach achieved higher accuracies (26.2% for ten letters, 46.7% for five letters) compared to direct EEG classification (23.1% and 39.0%).
• Hand kinematics were reconstructed with significant correlation (0.10-0.57).
• Written letters significantly influenced low-frequency EEG components, particularly in central and occipital channels.

Conclusions:

• Non-invasive neural signals can be used to extract handwritten letters.
• The proposed two-step approach enhances classification performance.
• Movement speed is a key kinematic feature for decoding handwriting from EEG.