Classification and Automated Interpretation of Spinal Posture Data Using a Pathology-Independent Classifier and...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Spinal Biomechanics

Background:

• Current clinical classification models are pathology-specific, limiting their diagnostic scope.
• There is a need for pathology-independent classifiers that offer interpretable predictions for spinal conditions.

Purpose of the Study:

• To develop a pathology-independent classifier for spinal posture analysis.
• To provide prediction probabilities and explanations for classification decisions.
• To evaluate the classifier's performance against traditional binary approaches.

Main Methods:

• Utilized spinal posture data from healthy subjects, patients with back pain, spinal fusion, osteoarthritis, and synthetic data.
• Employed a one-class support vector machine as the pathology-independent classifier.
• Applied Platt's method for probability transformation and Local Interpretable Model-Agnostic Explanations (LIME) for model interpretation.

Main Results:

• The classifier achieved the best performance in identifying spinal fusion.
• Distinguishing subjects with back pain from the healthy group proved challenging.
• The explainable AI tool effectively interpreted classification predictions.
• The proposed method showed no significant inferiority compared to binary classifiers.

Conclusions:

• The pathology-independent classifier offers a promising approach for objective spinal condition assessment.
• Interpretability of predictions is a key strength, aiding in therapy adaptation and monitoring.
• Future research should incorporate dynamic spinal data for enhanced model performance.
• The approach can support pre- and post-operative therapy management.