CNN-LSTM-Based EEG Eye Movement Classification for Assistive IoT Control in Locked-In Syndrome
Tonmoy, Shoaib Azad (2025-12-02)
CNN-LSTM-Based EEG Eye Movement Classification for Assistive IoT Control in Locked-In Syndrome
(02.12.2025)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe20251209116381
https://urn.fi/URN:NBN:fi-fe20251209116381
Tiivistelmä
Locked In Syndrome (LIS) is a rare neurological condition in which individuals remain fully conscious but experience near total paralysis, often retaining only limited eye movement. For such patients, even simple interaction with their surroundings can be profoundly challenging. This thesis investigates the feasibility of a low cost, non invasive brain–computer interface (BCI) capable of translating voluntary eye movements, detected through electroencephalography (EEG), into control commands for Internet of Things (IoT) devices.
A publicly available consumer grade EEG dataset was utilised, focusing on saccadic eye movement paradigms of varying complexity. The Muse S2 headset, with four frontal and temporal electrodes, provided the recording framework. The preprocessed EEG data provided by the dataset were directly segmented using a sliding window approach for feature extraction and model training. Two feature extraction strategies were considered: handcrafted statistical and spectral features for classical machine learning models, and end to end feature learning via a hybrid Convolutional Neural Network and Long Short Term Memory (CNN+LSTM) architecture.
Comparative experiments demonstrated that the CNN+LSTM model achieved the highest performance, with accuracies of 75.9% on structured Level 1 Saccades data and 69.0% on the more complex Level 2 Saccades data using a four class command configuration. The trained model was integrated into a simulated IoT environment, enabling eye movement driven control of common household devices such as lights, fans, and televisions.
The results confirm that reliable eye movement classification is feasible with consumer grade EEG and deep learning methods, offering a practical pathway toward accessible assistive technology for individuals with severe motor impairments. The proposed system provides a foundation for future work involving real time deployment, expanded device control, and integration with user specific calibration to enhance accuracy and usability in real world settings.
A publicly available consumer grade EEG dataset was utilised, focusing on saccadic eye movement paradigms of varying complexity. The Muse S2 headset, with four frontal and temporal electrodes, provided the recording framework. The preprocessed EEG data provided by the dataset were directly segmented using a sliding window approach for feature extraction and model training. Two feature extraction strategies were considered: handcrafted statistical and spectral features for classical machine learning models, and end to end feature learning via a hybrid Convolutional Neural Network and Long Short Term Memory (CNN+LSTM) architecture.
Comparative experiments demonstrated that the CNN+LSTM model achieved the highest performance, with accuracies of 75.9% on structured Level 1 Saccades data and 69.0% on the more complex Level 2 Saccades data using a four class command configuration. The trained model was integrated into a simulated IoT environment, enabling eye movement driven control of common household devices such as lights, fans, and televisions.
The results confirm that reliable eye movement classification is feasible with consumer grade EEG and deep learning methods, offering a practical pathway toward accessible assistive technology for individuals with severe motor impairments. The proposed system provides a foundation for future work involving real time deployment, expanded device control, and integration with user specific calibration to enhance accuracy and usability in real world settings.