A Hierarchical Machine Learning Solution for the Non-Invasive Diagnostic of Autonomic Dysreflexia

Abstract

More than half of patients with high spinal cord injury (SCI) suffer from episodes of autonomic dysreflexia (AD), a condition that can lead to lethal situations, such as cerebral haemorrhage, if not treated correctly. Clinicians assess AD using clinical variables obtained from the patient’s history and physiological variables obtained invasively and non-invasively. This work aims to design a machine learning-based system to assist in the initial diagnosis of AD. For this purpose, 29 patients with SCI participated in a test at Cruces University Hospital in which data were collected using both invasive and non-invasive methods. The system proposed in this article is based on a two-level hierarchical classification to diagnose AD and only uses 35 features extracted from the non-invasive stages of the experiment (clinical and physiological features). The system achieved a 93.10% accuracy with a zero false negative rate for the class of having the disease, an essential condition for treating patients according to medical criteria.