Machine learning models for predicting treatment response in infantile epilepsies

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Epilepsy is a common neurological disorder posing significant healthcare challenges.
• Machine learning (ML) offers versatile applications in healthcare, including diagnostics and prognosis.
• Predicting drug treatment outcomes in epilepsy is crucial for effective patient management.

Purpose of the Study:

• To compare 11 machine learning models for predicting drug treatment outcomes in pediatric epilepsy.
• To identify the optimal ML model for detecting drug-resistant epilepsy in a pediatric cohort.
• To evaluate the efficacy of ML in enhancing diagnostic accuracy for epilepsy treatment response.

Main Methods:

• Evaluated 229 pediatric patients (aged 1-24 months) diagnosed with epilepsy.
• Applied 11 machine learning techniques, including Support Vector Machine (SVM), Decision Trees, and Neural Networks.
• Utilized chi-square feature selection and performance metrics to assess anti-seizure medicine response.

Main Results:

• The Support Vector Machine (SVM) algorithm demonstrated high effectiveness in identifying drug-resistant epilepsy.
• SVM achieved the highest area under the curve (0.9934) and a test accuracy of 97.06%.
• The study included 229 pediatric patients, with a balanced gender distribution.

Conclusions:

• The Support Vector Machine (SVM) algorithm is highly effective for predicting drug-resistant epilepsy in pediatric patients.
• Findings suggest SVM can guide early referral to non-medical treatments like epilepsy surgery or ketogenic diets.
• A multidisciplinary approach is recommended for managing pediatric epilepsy, informed by advanced diagnostic tools.