Machine learning-based assessment of seizure risk predictors in myelomeningocele patients: A single-center retrospective cohort study

TL;DR

This study uses machine learning to predict seizure risk in patients with myelomeningocele based on factors like shunt history and imaging findings.

Contribution

The novel contribution is the application of machine learning models to identify seizure risk predictors in myelomeningocele patients.

Findings

ML models predicted seizure risk with 86–92% accuracy and AUC of 0.764–0.865.

Key predictors included gestational age, shunt history, and corpus callosum dysgenesis.

Imaging findings and shunt infection history were significant in predicting seizures.

Abstract

Myelomeningocele (MMC) is a severe congenital malformation of the CNS (central nervous system) that often leads to seizures due to factors such as shunt complications and hydrocephalus. This study aims to develop a machine learning model to predict the likelihood of seizures in MMC patients by analyzing various predictors.

This retrospective study involved 103 MMC patients. Factors such as demographics, MMC location, shunt history, and imaging were analyzed using the random forest classifier, the support vector classifier, and logistic regression. Model performance was assessed through bootstrap estimates, cross-validation, classification reports, and area under the curve (AUC).

Of the evaluated patients, 11 experienced seizures. The key influencing factors included gestational age, sacral location, hydrocephalus, shunt history, and corpus callosum dysgenesis. Machine learning (ML) models predicted seizure risk with an accuracy of 86–92% and an AUC ranging from 0.764 to 0.865. Significant predictors were imaging findings, shunt infection history, and gestational age.

ML models effectively predict seizure risk in MMC patients, with certain variables showing strong associations and significant impact.

Figures