Effective automated prediction of vertebral column pathologies based on logistic model tree with SMOTE preprocessing

Area of Science:

• Spinal Biomechanics and Medical Imaging Analysis
• Machine Learning in Healthcare
• Computational Pathology

Background:

• Vertebral column pathologies pose significant diagnostic challenges.
• Accurate and efficient detection of spinal conditions is crucial for patient management.
• Existing diagnostic methods may lack automation and comprehensive analysis.

Purpose of the Study:

• To develop an automated system for accurate recognition of vertebral column pathologies.
• To evaluate the efficacy of a Logistic Model Tree (LMT) based approach.
• To compare the performance of LMT with other machine learning algorithms for spinal pathology detection.

Main Methods:

• Utilized six biomechanical measures: pelvic incidence, pelvic tilt, lumbar lordosis angle, sacral slope, pelvic radius, and grade of spondylolisthesis.
• Employed a two-phase classification model: data preprocessing with Synthetic Minority Over-sampling Technique (SMOTE) followed by Logistic Model Tree (LMT) classification.
• Validated the system using 10-fold cross-validation on clinical records of 310 patients.

Main Results:

• Achieved 89.73% accuracy in computer-based automatic detection of vertebral column pathologies.
• Obtained an Area Under Curve (AUC) of 0.964, indicating strong diagnostic performance.
• Presented a comparative analysis of vertebral column data using various machine learning algorithms.

Conclusions:

• The developed LMT-based automation system demonstrates high accuracy and AUC for detecting vertebral column pathologies.
• SMOTE preprocessing enhances the performance of the LMT classifier in this context.
• The study provides a robust, validated method for computer-aided diagnosis of spinal conditions.