Prediction of functional assessment by ACL injury experts using markerless motion capture and machine learning through a multiplanar single-leg plyometric test

• Grant number: 25/12112-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: September 01, 2025
• End date: August 31, 2026
• Field of knowledge: Health Sciences - Physiotherapy and Occupational Therapy
• Principal Investigator: Paulo Roberto Pereira Santiago
• Grantee: Anna Carolina Magalhães Leonel
• Host Institution: Escola de Educação Física e Esporte de Ribeirão Preto (EEFERP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil

Abstract

This project aims to apply machine learning models and computer vision techniques to estimate functional assessments made by specialists for patients with Anterior Cruciate Ligament (ACL) injuries, using kinematic data obtained from markerless tracking during the Multiplanar Unipodal Plyometric (CUBE) test. Three GoPro Hero 10 cameras will capture movements in multiple planes, and the MediaPipe framework will extract relevant kinematic features associated with ACL injury risk factors. Specialists will assign global functional scores (1-5) to patients, based on video analysis and their prior clinical knowledge. Inter-rater reliability will be evaluated using Cohen's Kappa index. The resulting data will be used to train and evaluate multiple supervised machine learning models (Extreme Gradient Boosting, Random Forest, Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Gaussian Naive Bayes, Multilayer Perceptron), optimized with Grid Search and evaluated by stratified cross-validation. Model performance will be assessed by balanced accuracy, precision, recall, and F1 Score. Statistical analyses, performed in Python 3.12.11, will include normality (Shapiro-Wilk), homogeneity (Levene), group comparisons by one-way ANOVA or Kruskal-Wallis, appropriate post-hoc tests, and effect size estimation (Cohen's d). The ultimate goal is to develop an automated, objective, and reliable tool to support functional rehabilitation and clinical decision-making for post-ACL reconstruction patients and inidicates the best machine learning models for this task.