Abstract WP93: An Internet Of Things Based Real Time Location System With Machine Learning For Coordination Of Acute Stroke Endovascular Intervention: A Proof-of-concept Study

Abstract

Introduction: The delivery of mechanical thrombectomy in acute stroke can be challenging because it requires coordination between patient travel and multiple stroke team members across different locations in the hospital. A Real Time Location System (RTLS) can facilitate a streamlined workflow by determining a patient’s location, but the required infrastructure can be costly to set up. We investigated the feasibility of a machine learning based RTLS using an Internet Of Things (IOT) device prototype to coordinate acute stroke endovascular intervention. Methods: In this proof-of-concept study, study investigators programmed an IOT device to collect WiFi signals from different hospital zones relevant to the mechanical thrombectomy workflow. Different machine learning (ML) algorithms were trained on location labelled training WiFi signals to predict location. The accuracies of the different ML algorithms to predict unlabeled testing WiFi signals collected on a separate day were measured, in percentages. Results: The training and testing datasets contained 1,179 and 293 samples respectively, with 2,516 and 2,618 unique WiFi signals detected in each dataset. The trained neural network, random forest, support vector machine, K-nearest neighbors, decision tree and ensemble classifiers achieved accuracies of 96.6%, 96.2%, 97.3%, 97.3%, 83.6% and 97.3% on the testing dataset respectively. Additionally, the neural network model produced estimates of uncertainty, which were significantly associated with the correctness of predictions based on logistic regression analysis (p < 0.001). Conclusions: An IOT based real time location system with machine learning can accurately predict locations within the hospital, demonstrating immense potential for the future applications of streamlining delivery of acute stroke endovascular intervention.