Machine Learning Prediction of Autonomic Nervous System Dysfunction Using Multimodal Physiological Signals

Abstract

Background: This study was conducted to assess the Machine Learning Prediction of Autonomic Nervous System Dysfunction Using Multimodal Physiological Signals.

Material and methods: This research involved the recruitment and enrollment of 50 healthy, able-bodied individuals aged between 18 and 60 years, all with a BMI of less than 30. The average age (±SD) of the participants was 25.9 years, comprising sixteen males and five females. The average BMI recorded was 23.8. The exclusion criteria included: a history of cardiac arrhythmia, coronary artery disease, autoimmune disorders, chronic inflammatory diseases, anemia, malignancies, depression, neurological disorders, diabetes mellitus, renal diseases, dementia, psychiatric conditions including active psychosis, or any other chronic medical issues, treatment with anti-cholinergic medications, current use of tobacco, nicotine, or other recreational drugs, pre-existing neurological conditions, pregnancy, and the presence of implantable electronic devices. Participants were instructed to fast and avoid caffeine for a minimum of four hours before the testing commenced. Testing sessions occurred in a laboratory setting with shielded from external noise, light, or distractions, with an average humidity of 22% and average temperature of 22.1 °C. Lighting was set such that significant pupil changes were detected, averaging approximately 22 lm. Cardiovascular, pupil dilation, and EDA signals were synchronously recorded for each participant. The individual responses of calculated signals during each test were averaged to determine a response template.

Results: From all the physiological modalities we monitored, participants’ cardiovascular measures (HR, MAP, and RMSSD) registered average responses with a peak above 1σ of their baseline, thus deemed significant responses, while the pupil dilation and EDA measures did not show a consistent, significant response and were discarded from further processing and feature extraction.

Conclusion: These findings illustrate the reliability and responsiveness of an algorithmic technique for extracting multimodal responses from conventional assessments. This innovative approach to quantifying autonomic nervous system (ANS) function can facilitate early diagnosis, track disease progression, or assess treatment efficacy.