Real-Time Hand Movement Detection using a Custom-Built EEG System

Spinal cord injury causes great disruption of voluntary hand movement, with consequent restrictions in functional independence. This work presents a real-time EEG-based detection system for hand opening and closing movements using a custom-built low-cost acquisition device. The proposed system includes eight EEG channels over motor-related cortical regions, amplified by AD620 instrumentation amplifiers, filtered through analog stages, and digitized with a 16-bit ADS1115 converter. EEG signals were recorded from healthy volunteers who continuously performed forced hand-opening and hand-closing tasks to capture clear cortical patterns associated with both states. Preprocessing included the reduction of artifacts, normalization, and feature selection based on signal variance to enhance the signal. Afterwards, a transformer-based deep learning model was developed to identify the hand state in real time with high accuracy and good temporal stability of the results. The results demonstrate that the combination of lightweight hardware with advanced neural network models allows for reliable detection of motor intent from non-invasive EEG signals. The results showed that the system achieved an accuracy of 97.78% with a latency less than 200 ms. This approach leads to low-cost neurotechnology for real-time assistive applications, pointing out the potentiality of its adaptation for restoring hand functionality in individuals with SCI.