Optimizing prediction accuracy for early recurrent lumbar disc herniation with a directional mutation-guided SVM...

Area of Science:

• Spinal Surgery
• Machine Learning
• Medical Diagnostics

Background:

• Percutaneous endoscopic lumbar discectomy (PELD) is a common minimally invasive treatment for lumbar disc herniation.
• Predicting early recurrence of lumbar disc herniation (rLDH) remains a challenge.
• Current machine learning models require improved feature selection for rLDH prediction.

Purpose of the Study:

• To develop an enhanced classification model for predicting early recurrent lumbar disc herniation (rLDH).
• To identify key risk factors contributing to rLDH using an integrated machine learning approach.
• To improve the accuracy and interpretability of rLDH prediction models.

Main Methods:

• Utilized a wrapper feature selection method integrating an enhanced bat algorithm (BDGBA) with a support vector machine (SVM).
• BDGBA employed directional mutation and guidance-based strategies to optimize feature subsets.
• SVM acted as the classifier, with its prediction accuracy serving as the fitness function for feature evaluation.

Main Results:

• The proposed model achieved 93.49% accuracy and 88.33% sensitivity in predicting rLDH.
• Key predictive factors identified include herniated disk level, Modic changes, disk height, length, and width.
• The model demonstrated effectiveness in real-world dataset prediction experiments.

Conclusions:

• The integrated BDGBA-SVM model offers an effective auxiliary diagnostic tool for predicting rLDH.
• The identified key factors provide valuable insights into the causes of early recurrence after PELD.
• This approach enhances the prediction accuracy and interpretability of rLDH diagnostic models.