Detecting and Eliminating Cardiac Artifact from Endovascular EEG Signals

Paralysis is a debilitating condition that affects more than 5.4 million people in the U.S. In severe cases, the paralyzed patient is incapable of communication. Restoring this communication is a primary goal of caretakers and is critical to improving the patient's quality of life. Brain-computer interfaces (BCIs) that directly access signals from the motor cortex are a promising method of circumventing the condition causing paralysis, typically using machine learning (ML) to predict motor intent from brain signals. However, BCIs are highly invasive and subjects have primarily been limited to patients with mild to moderate paralysis. The Stentrode is a novel technology that records electroencephalographic (EEG) signals via an electrode array placed endovascularly in the superior sagittal sinus. The first clinical trials of this technology aim to enable digital communication for severely paralyzed patients, translating brain signals from attempted movements into computer control inputs like mouse clicks. However, recordings of EEG are often contaminated with artifacts, including biopotentials arising from other excitable tissues, such as the heart and skeletal muscle. This study characterizes the electrocardiographic (ECG) artifact detected in the Stentrode recordings and proposes an automated Independent Component Analysis (ICA) method for removing this artifact. We compare the effectiveness of this method to previous methods for removal. Quantifying and eliminating the cardiac artifact is critical to accurately decode signals from the motor cortex and restore patients' ability to communicate.