Creative Commons source reading · Machine learning for performance
Making Sense of Shoulder Exercise: Measuring the Accuracy of an Artificial Intelligence Model to Classify Shoulder Exercise via Wearable Sensors Among People With and Without Rotator Cuff Tendinopathy
Josh Naunton; Yanran Jiang; Rodrigo Bini; Dawson Kidgell; Kim Bennell; Terry Haines; Dana Kulić; Peter Malliaras
Originally published in Eur J Sport Sci · 2026-05-01
From the article
Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.
Abstract
ABSTRACTThis study aimed to compare the accuracy of machine learning classification for three commonly prescribed shoulder exercises in people with and without rotator cuff tendinopathy. Eighteen participants with rotator cuff tendinopathy (mean age 54.2, SD 13.2; 50% female), followed by eighteen matched controls completed a laboratory‐based shoulder strength testing protocol. Three exercises were performed (shoulder press, lateral raise and bent over row) while wearing three inertial measurement (IMU) sensors (Axivity, Ax6 ‐ 3 axis accelerometry and gyroscope at 100 Hz and 1000°/sec respectively) positioned on the wrist arm and trunk. Data were analysed and accuracy was compared between common machine learning algorithms for those with rotator cuff tendinopathy and healthy matched controls in a subject‐dependent and…
Results
Participant characteristics and demographics are summarised below in Table 2. Waist circumference measure was the only baseline outcome with a statistically significant difference between the groups (p value = 0.04, two sample t‐test). For the 1RM strength test results, the lateral shoulder raise showed a statistically significant difference between groups (see Table 3).
Discussion
This study demonstrates the accuracy of artificial intelligence, utilising machine learning to automatically classify shoulder exercise for both patients with rotator cuff tendinopathy and a matched pain free cohort. Classification accuracy is similar between a subject‐dependent model and subject‐independent model. Overall, accuracy was greater for the entire cohort with a subject‐independent model which may reflect the larger data pool available. Interestingly, utilising a single wrist based sensor achieves similar results to the 3‐sensor system (within 3%–5% accuracy) for the entire cohort using a subject‐independent analysis. The ability to automatically detect and recognise exercise via wrist worn wearable technology presents a promising opportunity to objectively provide clinically meaningful data to clinicians and researchers,…
Editorial reading guide
- What data created and tested the model, and was evaluation independent?
- What is the cost of a false positive, false negative or out-of-distribution decision?
