Classification of low back pain from dynamic motion characteristics using an artificial neural network

Area of Science:

• Biomechanics and Computational Intelligence
• Spinal Disorder Diagnostics

Background:

• Evaluating lower back pain is challenging, with imaging techniques often being costly and ineffective.
• A novel approach utilizes dynamic motion features (shape, velocity, symmetry) and neural networks for lower back pain assessment.

Purpose of the Study:

• To identify specific trunk motion characteristics linked to various spinal disorder categories.
• To assess the efficacy of a neural network system in differentiating these motion patterns.

Main Methods:

• Collected dynamic motion data from 183 subjects using a triaxial goniometer.
• Extracted movement features for a two-stage neural network classifier (radial basis function).
• Compared classifier output against Québec Task Force pain classifications, evaluating linear and nonlinear techniques.

Main Results:

• The neural network system successfully classified low back pain based on motion characteristics.
• Achieved up to 85% accuracy on validation data, demonstrating strong predictive capability.

Conclusions:

• A neural network model utilizing kinematic data provides an effective method for classifying low back pain.
• This technology has the potential to significantly enhance the clinical management of individuals with low back pain.