Page 139 - AIH-1-2

Artificial Intelligence in Health Movement detection with sensors and AI

level but also pose detrimental challenges to hospital has aided in predicting the evolution of mild cognitive
administrations and insurance companies, as they not impairment to Alzheimer’s disease. In recent years,
12
only lead to injuries and increased risk of fatal events machine learning has seen growing use in the healthcare
but simultaneously extend hospital stays and inflate industry with the objective of enhancing patient results
medical care costs. According to the Center for Disease and fostering more effective and tailored care practices. 13,14
Control’s National Center for Injury Prevention and In this study, PyCaret was used to classify data surrounding
Control, unintended injuries are responsible for more both simulated patient falls and simulated patient seizures,
years of potential life lost than any other cause of death; specifically classifying data relating to six particular
among the reported 3.4 million unintended injuries, motions: breathing, seizures, rolling to the right side,
72,000 are attributed to falls. While inpatient hospital rolling to the left side, rolling off the bed from the left, and
2
falls are objectively viewed as preventable events thus far, rolling off the bed from the right. This study serves as an
there remains a dearth of effective preventive measures. innovative approach to fall prevention that has not been
3,4
Current methods include providing patients with previously implemented in hospitals. Using the Movella
educational videos on fall prevention, deploying various Xsens motion sensors to continuously gather continuous
forms of bed alarms, and employing video monitoring data on patient movements and employing machine
in patient rooms. Falls occur for a variety of reasons. learning algorithms to classify said data, trigger alert
5,6
Accidental falls occur when patients slip, trip, or encounter systems for hospital staff may be developed. The goal is to
other environmental factors. Anticipated physiological prevent adverse hospital events such as falls and hospital-
falls can be best described as falls experienced by patients onset seizures, thereby leading to better patient outcomes
predisposed to falling, influenced by factors such as and improved patient safety.
previously recorded falls, an inaccurate self-assessment
of capabilities, the presence of intravenous lines or saline This study directly tackles the significant challenge
locks, or the use of an ambulatory aid. Anticipated posed by inpatient falls and hospital-onset seizures,
7,8
physiological falls constitute the majority of inpatient falls, occurrences that not only jeopardize patient safety but also
while unanticipated physiological falls are less frequent and impose considerable costs on healthcare systems. Despite
unpredictable. This study aims to address the longstanding existing preventative measures, these events remain
8
challenge of hospital patient falls by presenting a solution. a concern. In response, this study introduces a novel
approach by utilizing the Movella Xsens sensors alongside
Hospital-onset seizures, defined as seizures occurring the PyCaret machine learning algorithm to predict and
in hospitalized patients not admitted for seizure-related potentially prevent such incidents. The sensor device
incidents and lacking a history of seizures, represent jarring detects real-time motions, while the PyCaret algorithm
occurrences often associated with extended hospital stays classifies these movements to recognize patterns associated
and heightened medical care requirements. A previous with risk events. This integrated approach was tested
9
study investigating hospital-onset seizures identified using mannequins, demonstrating an 89% accuracy in
218 patients, revealing that 33% experienced generalized movement detection. The findings suggest the potential of
tonic-clonic seizures, while metabolic derangements this technology to serve as an effective alert system, thereby
accounted for 25% of the remaining cases. In addition, the advancing patient safety by enabling timely interventions
9
study discovered a higher incidence of mortality among by medical staff.
patients experiencing hospital-onset seizures compared to
those with preexisting histories of seizures, with rates of 2. Materials and methods
19% and 5%, respectively. Thus, hospital-onset seizures
9
typically manifest as new-onset and often recur, coinciding This section provides an overview of the experimental
with elevated mortality rates. The results gathered in this data collection process, operationalization of sensors and
study on patient seizures propose a potential novel safety movements to obtain relevant data, data preprocessing for
measure for early seizure detection and swift intervention. machine learning analysis, and utilization of the PyCaret
machine learning library to establish, analyze, and evaluate
PyCaret is a low-code, open-source machine learning
library within Python designed to streamline coding classification models. It emphasizes the metrics used to
determine the success and reliability of the models in
efforts while increasing the time available for analysis. Its predicting different types of patient movement, ultimately
application has extended to evaluating turnaround time, contributing to the study’s goal of improving patient safety
a critical performance indicator in medical diagnostic through early detection of fall or seizure events.
laboratories. In addition, PyCaret has demonstrated
10
promise in studies focusing on histological variants of In conducting this study, a methodology was utilized
bladder and urothelial carcinomas. Notably, PyCaret to replicate the movements associated with inpatient
11

Volume 1 Issue 2 (2024) 133 doi: 10.36922/aih.2790