Wavelet analysis reveals differential lower limb muscle activity patterns long after anterior cruciate ligament reconstruction

• Published in: Journal of biomechanics Vol. 133; p. 110957
• Main Authors: Zandiyeh, Payam, Parola, Lauren R., Fleming, Braden C., Beveridge, Jillian E.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.03.2022; Elsevier Limited
• Subjects: Activity patterns; Anterior cruciate ligament; Anterior Cruciate Ligament Injuries - surgery; Anterior Cruciate Ligament Reconstruction; Biomechanics; Demographics; Electromyography; Electromyography, wavelet analysis; Females; Force; Hamstring Muscles; Humans; Joint and ligament injuries; K-nearest neighbor; Knee; Learning algorithms; Ligaments; Lower Extremity - surgery; Machine learning; Motion capture; Muscle contraction; Muscle function; Muscle, Skeletal - physiology; Muscles; Neuromuscular function; Patients; Prospective Studies; Quadriceps muscle; Quadriceps Muscle - surgery; Skeletal muscle; Surgery; Wavelet Analysis; Anterior cruciate ligament reconstruction; Neuromuscular function; Electromyography, wavelet analysis; K-nearest neighbor
• ISSN: 0021-9290; 1873-2380
• DOI: 10.1016/j.jbiomech.2022.110957

The purpose of this study was to test whether differences in muscle activity patterns between anterior cruciate ligament-reconstructed patients (ACLR) and healthy controls could be detected 10 to 15 years post-surgery using a machine learning classification approach. Eleven ACLR subjects and 12 healthy controls were recruited from an ongoing prospective randomized clinical trial. Surface electromyography (EMG) signals were recorded from gastrocnemius medialis and lateralis, tibialis anterior, vastus medialis, rectus femoris, biceps femoris, and semitendinosus muscles. Muscle activity was analyzed using wavelet analysis and examined within four sub-phases of the hop test, as well as an average of the task as a whole. K-nearest neighbor machine learning combined with a leave-one-out validation was used to classify the muscle activity patterns as either ACLR or Control. When muscle activity was averaged across the whole hop task, activity patterns for all muscles except the tibialis anterior were identified as being different between the study cohorts. ACLR patients demonstrated continuous muscle activities that spanned take-off, airborne, and landing hop phases versus healthy controls who displayed timed and regulated islets of muscle activities specific to each hop phase. The most striking features were 25–50% greater relative quadriceps intensity and approximately 66% diminished biceps femoris intensity in ACLR patients. The current findings are in contrast to previous work using conventional co-contraction and muscle activation onset EMG measures of the same dataset, underscoring the sensitivity and potential of the wavelet approach coupled with machine learning to reveal meaningful adaptation strategies in this at-risk population.