Top Awards

Roche Diagnostics Best of Engineering, Robotics and Computer Sciences Award

Grand Award

3rd Place

Sponsored Awards

Department of Navy Naval Science Award

Comparing Different Machine Learning Approaches to Assess Frailty Among Geriatric Patients Using Upper-Extremity Function

Biochemical Science and Bioengineering
Ananya Munjal

Elyse Wexler

Frailty, or a general weakness of the body, is a beneficial indicator of patient disease progression and possible health outcomes. The Upper-Extremity Function (UEF) test calculates patient frailty by having subjects move their forearm while wearing sensors and running the data through a UEF software. This test has the capability to become the new clinical standard for calculating frailty, replacing the current laborious testing process, however its accuracy needs to be increased. To make the UEF test more accurate, we tested different machine learning models and extracted more complex features from the software data. Previously, the team extracted some features from UEF signals, categorized the older adult according to their frailty status, and classified them with an accuracy of 65%. To increase the accuracy, we determined frailty statuses of different subjects using the UEF software and then took parameters from the output data which we compared to the frailty status values using different machine learning models. The accuracy comes from how well these parameters are able to predict the frailty status. Long Short-Term Memory deep learning proved to be the most accurate with 75.7% accuracy. The best feature set of parameters was speed mean, flexion numbers, sample entropy, PSD, and range of motion variability. This application of machine learning will allow for a faster, more accurate way to assess frailty and the progress of patients, allowing the geriatric community to live longer, more fulfilling lives. This test has the potential to become the clinical standard for testing frailty.

Project presentation

View Project Presentation file

Lab journal excerpts

View Lab Journal file

Research paper

View Research Paper file