Bilateral upper-humerus IMU sensors identified lower left-right shoulder waveform cross-correlation in swimmers with moderate/severe shoulder pain history (0.78 vs. 0.85, p = 0.002), but machine learning classification of pain group was only modest (balanced accuracy ~70%), suggesting personalized approaches may be needed.
Key Findings
Results
Swimmers with none/mild shoulder pain history showed higher peak left-right shoulder cross-correlation across all axes than swimmers with moderate/severe pain history.
Peak cross-correlation was 0.85 in the none/mild group vs. 0.78 in the moderate/severe group (p = 0.002).
Participants were 34 NCAA Division 1 collegiate swimmers and water polo players.
Groups were defined by self-reported shoulder pain scores: moderate/severe (n = 22) and none/mild (n = 12).
IMU sensors were placed bilaterally on the upper humerus during a standardized swimming test.
Results
A machine learning model distinguished moderate/severe from none/mild shoulder pain history with modest accuracy.
The best-performing model reached a balanced accuracy of 70.6% (24/34 participants).
A balanced accuracy of 69.7% (95% CI 52.7-85.2%) was reported.
ROC AUC was 0.731 (95% CI 0.557-0.890).
Kinematic features were extracted and evaluated using statistical tests and machine learning classification.
Conclusions
The authors conclude that personalized machine learning approaches may be needed to improve individual-level classification of shoulder pain history.
Group-level classification performance was modest with wide confidence intervals.
Further investigation of intra-subject associations between pain history and bilateral waveform similarity is called for to understand the relationship and potential causal mechanisms.
Background
The study addresses a gap in objective, field-deployable methods for quantifying pain-related movement alterations in swimmers.
Shoulder injuries are described as one of the most common overuse problems in competitive swimming.
The study tested whether bilateral IMU sensors on the upper humerus can identify bilateral kinematic waveform similarities tied to shoulder pain history.
What This Means
This study asked whether small wearable motion sensors placed on both upper arms could pick up differences in how swimmers move if they have a history of shoulder pain. Thirty-four college swimmers and water polo players did a standardized swim test and were grouped by their self-reported shoulder pain: 22 with moderate/severe pain and 12 with none/mild pain.
Swimmers with little or no shoulder pain showed more similar movement patterns between their left and right shoulders (similarity score 0.85) than swimmers with moderate/severe pain (0.78). When the researchers used machine learning to classify swimmers into the two groups from the sensor data, it got about 70% balanced accuracy, with fairly wide uncertainty ranges.
This research suggests that left-right movement symmetry measured by wearable sensors may be related to shoulder pain history, and such sensors could eventually offer a practical way to monitor swimmers. However, the small sample, modest classification accuracy, and the fact that the study cannot show whether asymmetry causes pain or results from it mean more work is needed. The authors suggest following individual swimmers over time and using personalized models.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.