Title : Facilitating emotional regulation through brain-computer interfaces
Abstract:
Neurological and physiological disorders such as Autism Spectrum Disorder (ASD), ADHD, PTSD, and Bipolar Disorder uniquely impair emotional regulation, presenting complex challenges for therapeutic intervention. This research explores the potential of EEG-based Brain-Computer Interfaces (BCIs) as a unified yet personalized technological solution to assist individuals with such disorders in recognizing and managing their emotional states. Specifically, the study focuses on developing a novel neural network architecture capable of classifying emotional states from EEG signals, leveraging power spectral density analysis and data complexity augmentation through multiscale entropy and Gaussian noise simulation to mirror neurological disorder patterns. EEG emotion data was sourced from the OpenNeuro DENS dataset and strategically filtered to extract signals from electrode positions compatible with consumer-grade EEG devices. A comprehensive preprocessing pipeline—including bandpass filtering, standard score normalization, and oversampling via SMOTE—was implemented to prepare the dataset. Simulated disorder complexity was achieved by modulating entropy, allowing for realistic training conditions without direct patient data. To process the EEG signals, power spectral density values were transformed into 2D matrices suitable for CNN-based classification. Transfer learning models, including ResNet50-v2, Inception-v3, and MobileNet-v2, were trained and evaluated for binary and categorical emotional classification. Among these, ResNet50-v2 achieved the highest accuracy (96.4% binary classification) and robust performance in real-time conditions. The findings indicate a high potential for deploying lightweight, real-time EEG-BCI systems that deliver automated, responsive interventions—such as calming stimuli, guided breathing, or affirmations—based on detected emotional states. These systems may offer substantial improvement over traditional therapeutic approaches by enabling continuous, personalized emotional support with minimal external involvement. Future work includes collecting EEG data from individuals diagnosed with neurological disorders to increase sample diversity and model generalizability, as well as integrating advanced architectures such as attention-augmented RNN ensembles. This research represents a step forward in building accessible, adaptive BCI technologies for emotional regulation across diverse neurophysiological conditions.
