MRI machine learning model predicts nerve root sedimentation in lumbar stenosis: a prospective study

Area of Science:

• Radiology
• Spinal Imaging
• Biomedical Engineering

Background:

• Lumbar spinal canal stenosis (LSS) is a common condition causing nerve root compression.
• The nerve root sedimentation sign (SedSign) is an MRI finding in LSS, but its characteristics and predictors are not fully understood.

Purpose of the Study:

• To analyze MRI characteristics of SedSign in LSS.
• To develop a predictive risk model for SedSign occurrence in LSS patients.

Main Methods:

• 1,138 LSS cases analyzed, categorized by SedSign presence.
• Key MRI parameters (diameters, thicknesses, herniation, hypertrophy) and clinical data collected.
• Recursive feature elimination with cross-validation (RFECV) and machine learning (Random Forest, KNN, XGBoost) used for model development and evaluation.

Main Results:

• SedSign-positive LSS patients were older, with greater stenosis, increased epidural fat, disc herniation, ligamentum flavum hypertrophy, and high-intensity zones.
• RFECV identified eight predictive features: age, sex, dural sac dimensions, epidural fat grade, disc herniation, and ligamentum flavum hypertrophy.
• The SedSign8 model demonstrated high predictive performance (AUC=0.901, accuracy=83.6%).

Conclusions:

• Older age, increased stenosis, and altered dural sac/surrounding tissues are key factors for SedSign in LSS.
• The SedSign8 model offers a reliable tool for predicting SedSign risk in LSS patients.