Detection of Dysautonomia in Spinal Cord Injury Through Non-Invasive Multi-Modal Sensing and Machine Learning

Shruthi Suresh, Purdue University

Abstract

Dysautonomia is the dysfunction of the Autonomic Nervous System (ANS) that frequently occurs in individuals with spinal cord injuries (SCI), stroke, diabetes, or Parkinson’s disease. Dysautonomia after SCI that results in tetraplegia most commonly presents as autonomic dysreflexia (AD). AD can be triggered by different stimuli below the level of injury resulting in paroxysmal hypertension. If not properly managed, AD can have severe clinical consequences, leading to stroke and potentially death. AD is currently detected in-clinic through continuous monitoring of blood pressure using a cuff-based system. However, existing techniques are time-consuming, obtrusive, lack automated detection capabilities, and have low temporal resolution. Thus, a wearable diagnostic tool was developed that could detect the onset of AD using non-invasive physiological sensors through repeatable machine learning and data science techniques. This work presents a novel, multimodal system that can quantitatively characterize and distinguish unique signatures of AD. We used rodent models of SCI to detect finer temporal changes in the sympathetic and parasympathetic branches of the ANS due to AD. Signal processing and feature selection techniques were used to determine five features which were most significant to characterizing AD. This allowed us to characterize a concomitant increase in sympathetic activity followed by an increase in vagal activity during the onset of AD. Additionally, we used the unique signature to train a neural network to detect the onset of AD with an accuracy of 93.4%. We developed a model that can distinguish between reactions of sympathetic hyperactivity due to different stimulus triggers above and below the level of injury. The system could serve as a complementary tool to the clinically accepted gold standard of determining AD using solely blood pressure, providing a method for universally detecting the onset of AD and discriminating the different triggers for sympathetic stress for improved management of AD in individuals with SCI.

Degree

Ph.D.

Advisors

Duerstock, Purdue University.

Subject Area

Physiology|Artificial intelligence|Electrical engineering|Medicine|Morphology|Neurosciences|Public health|Statistics

COinS