Machine Learning-Driven Improvements in HRV Artifact Correction for Psychosis Prediction in the Schizophrenia Spectrum

Exploring the complex interplay between the human mind and heart, our study delves into the role of Heart Rate Variability (HRV) as a non-invasive marker of psychosis, influenced by the Central Nervous System (CNS). In the current landscape of mental health research, the emphasis extends beyond merely diagnosing and unveiling biological markers. It encompasses predicting the flow of psychotic manifestations, leveraging biomarkers that transcend mere diagnostic labels. HRV, derived from the analysis of beat-to-beat intervals obtained through advanced smartwatch technology, offers a window into the delicate balance of our autonomic cardiac regulation. However, these beat-to-beat intervals are often fraught with anomalies or missing values, physiological like ectopic beats, and technical, possibly stemming from a subject's motion or reaction to external stimuli. Addressing these aberrations is vital, as they can compromise the integrity of HRV assessments. Our research tackles challenges such as ectopic beats and inconsistencies in beat-to-beat intervals, which are critical to ensuring the accuracy of HRV measurements. We employ a structured three-stage decision-making process, utilizing four machine learning algorithms to systematically evaluate and refine our approach. This process includes selecting the best dataset that captures sleep and awake states, identifying the most effective interpolation method, and choosing the best ectopic beat correction technique. This step-by-step comparison helps in selecting the optimal combination of dataset, interpolation, and ectopic beat correction for our study. Through this approach, we aim to enhance the reliability of HRV as a tool for mental health research, providing insights into the autonomic regulation in psychiatric conditions.