Please login to view this media
- Talk
- 15/06/2021
- Canada
Machine Learning Algorithm for Evaluating Limb Symmetry Index Cut-Off Thresholds in ACL Injured Paediatric Patients
Description
In this presentation, Nicholas Romanchuk, PhD candidate at the University of Ottawa, discusses a study exploring the use of machine learning algorithms to establish the optimal limb symmetry index cutoff for classifying pediatric patients with ACL injuries. Nicolas explains that current practice lacks consensus on the best threshold for assessing readiness to return to activity, as assessments typically rely on limb symmetry indices, which afford a range of 75% to 95%.
He notes a systematic review he conducted highlighted a knowledge gap regarding these thresholds and sets out to determine a more precise cutoff. He collected data from 42 ACL-injured pediatric patients and 69 uninjured controls, all of whom underwent conventional return-to-activity tests, yielding various limb symmetry indexes. The findings indicate that uninjured participants generally had better symmetry scores in most tests.
To pinpoint optimal thresholds, Romanchuk employs two machine learning methods: a decision tree algorithm and a categorical transformation of data based on different cutoff values (80% as an example). Results suggest that a threshold between 84% and 86% effectively distinguishes between injured and uninjured populations, a conclusion supported by both analysis methods.
He emphasizes the implications of these findings and notes that they corroborate ongoing practices of utilizing 85% or 90% thresholds for pediatric classifications. Future research may explore specific cutoffs for different activities and develop improved classification methods beyond limb symmetry indexes. Romanchuk concludes by expressing gratitude for the audience's attention and looking forward to discussions at the upcoming North American Congress on Biomechanics.