Muscle Fat and Volume Differences in People With Hip-Related Pain Compared With Controls: A Machine Learning...

Area of Science:

• Orthopedics
• Sports Medicine
• Medical Imaging

Background:

• Hip-related pain (HRP) affects active adults, impacting quality of life and physical function.
• Conditions like femoroacetabular impingement syndrome and labral tears cause HRP.
• Hip muscle dysfunction is noted in HRP, but evidence in younger populations is limited.

Purpose of the Study:

• Compare hip muscle volume and intramuscular fat infiltrate (MFI) in athletes with and without HRP.
• Assess the reliability and accuracy of automated machine learning (ML) for hip muscle segmentation.
• Compare ML segmentation to human-generated segmentation.

Main Methods:

• Cross-sectional study of sub-elite football players with (n=180) and without (n=48) HRP.
• MRI assessed gluteus maximus, medius, minimus, and tensor fascia latae muscle volume and MFI.
• Linear regression analyzed muscle volume associations, controlling for covariates.
• Convolutional Neural Network (CNN) ML compared to manual segmentation in a subset (n=52).

Main Results:

• Significant differences in gluteus medius and tensor fascia latae muscle volumes between groups.
• No significant differences in gluteus maximus or minimus muscle volumes.
• CNN achieved high segmentation accuracy (>0.900) and reliability (>0.900), outperforming manual raters.

Conclusions:

• Increased gluteus medius and tensor fascia latae muscle volumes in HRP may relate to hypertrophy, potentially enhancing muscular efficiency.
• Automated CNN segmentation is efficient and highly reliable for assessing hip muscle morphology.
• ML offers a promising, accurate alternative to manual segmentation in clinical and research settings.