A Hybrid Ensemble Learning Framework for Predicting Lumbar Disc Herniation Recurrence: Integrating Supervised Models,...

Area of Science:

• Neurosurgery
• Machine Learning
• Clinical Data Analysis

Background:

• Lumbar disc herniation (LDH) recurrence poses a significant clinical challenge.
• Existing predictive tools for LDH recurrence are limited.
• Class imbalance and low signal-to-noise ratio complicate algorithmic prediction of LDH recurrence.

Purpose of the Study:

• To develop and evaluate a hybrid machine learning framework for predicting LDH recurrence.
• To integrate supervised and unsupervised learning methods with threshold tuning.
• To utilize routine clinical data for enhanced prediction of LDH recurrence.

Main Methods:

• A dataset of 977 patients was analyzed.
• A hybrid framework combining deep neural networks, random forest, and autoencoders was implemented.
• Ensemble stacking and sensitivity-focused threshold tuning were employed for optimization.

Main Results:

• Baseline models achieved high accuracy but zero sensitivity for recurrence.
• The proposed ensemble model demonstrated 100% internal recall at a threshold of 0.05.
• Key predictors identified included hospital stay, L4-L5 herniation, obesity, and hypertension.
• External validation revealed a significant drop in performance to 0% recall, indicating poor generalization.

Conclusions:

• The hybrid ensemble model effectively detects rare LDH recurrence cases under internal validation.
• Poor external performance underscores the difficulties in modeling rare clinical events.
• Future research should focus on external validation, longitudinal data, and model interpretability for clinical application.