Knee vibrations may signal cartilage damage

Machine-learning models distinguished healthy knees from those with osteoarthritis-related cartilage damage, but age and BMI performed as well as or better than the vibration-based models.

By
Andrea Surnit
October 6, 2026
3 min

MDSpire News

Overview

Vibroarthrography combined with machine learning shows potential as a noninvasive method for identifying cartilage damage in osteoarthritis patients, but demographic confounding factors complicate the interpretation of its diagnostic performance.

Background

Knee osteoarthritis is a prevalent degenerative joint disorder that leads to significant pain and functional limitations. Accurate detection of cartilage damage is crucial for timely intervention and management. Current diagnostic methods primarily rely on clinical assessment and conventional imaging.

Data Highlights

Condition	Classifier	AUC	Accuracy	Sensitivity	Specificity
Closed Kinetic Chain	Random Forest	0.917	79%	69%	88%
Closed Kinetic Chain	XGBoost	0.912	79%	69%	88%
Closed Kinetic Chain	AdaBoost	0.910	79%	69%	88%
Open Kinetic Chain	Random Forest	0.857	78%	68%	86%
Open Kinetic Chain	XGBoost	0.846	78%	68%	86%

Key Findings

Vibroarthrography (VAG) signals were recorded from 97 participants, including 49 patients with symptomatic knee osteoarthritis.
Random Forest classifier achieved the highest AUC of 0.917 under closed kinetic chain conditions.
Demographic differences, such as age and BMI, were significant confounding factors affecting diagnostic performance.
Simple signal characteristics like root-mean-square and crest factor were among the most informative features for classification.
Further validation across different sensor types and configurations is necessary.