NUST Institutions Library Catalogue

Low back pain (LBP) is one of the most prevalent musculoskeletal disorders worldwide,
with diagnosis often relying on subjective evaluation rather than objective physiological
measures. This thesis introduces a data-driven framework for quantitative assessment of
LBP using high-density surface electromyography (HD-sEMG). As a non-invasive
technique, HD-sEMG provides insight into spinal neuromuscular behavior, enabling
spatial and temporal characterization of muscle activity patterns associated with
dysfunction. The study recruited 39 participants, divided into three groups that are healthy,
sub-clinical, and LBP based on chiropractic evaluation. In the first study, machine learning
classifiers including Support Vector Machine (SVM), eXtreme Gradient Boosting
(XGBoost), and Artificial Neural Network (ANN) were trained on time and frequencydomain features to discriminate between groups. The SVM model achieved the highest
accuracy, effectively distinguishing subtle neuromuscular differences between healthy and
dysfunctional subjects. In the second study, a regression-based framework was developed
to predict vertebral joint dysfunction scores (C1-Sacral) derived from chiropractic
assessment. ANN and Convolutional Neural Network (CNN) models were trained under a
CORAL (Consistent Rank Logits) ordinal regression framework, preserving the ordinal
nature of dysfunction severity. The ANN model demonstrated superior predictive
performance, capturing non-linear relationships between HD-sEMG activity and graded
dysfunction levels. Overall, this research bridges the gap between clinical assessment and
computational diagnostics, showing that HD-sEMG signatures can objectively quantify
spinal dysfunction and support data-driven LBP diagnosis. The proposed framework
establishes a foundation for personalized rehabilitation, automated dysfunction mapping,
and AI-assisted musculoskeletal diagnostics, advancing the integration of biomedical
signal processing with clinical neurophysiology.