Machine learning algorithms predict canine structural epilepsy with high accuracy.

Area of Science:

• Veterinary neurology
• Machine learning in animal health
• Canine epilepsy diagnostics

Background:

• Clinical reasoning in veterinary medicine often relies on experience and existing literature.
• Decision-making for individual patient cases, particularly predicting underlying pathologies in seizuring dogs, lacks robust scientific methods.
• There is a need for advanced tools to aid in the diagnosis of structural epilepsy in dogs.

Purpose of the Study:

• To apply machine learning algorithms to predict the risk of structural epilepsy in dogs experiencing seizures.
• To develop a data-driven approach for diagnosing canine epilepsy.
• To enhance clinical decision-making for seizuring dogs.

Main Methods:

• Retrospective and prospective inclusion of dogs with seizure history.
• Utilized Bayesian Network and Random Forest algorithms for analysis.
• Applied feature selection methods including Permutation Importance, Forward Selection, Random Selection, and Expert Opinion to identify key predictive factors.

Main Results:

• A total of 328 dogs were analyzed, with 33.2% diagnosed with structural epilepsy.
• The Bayesian Network with Random Feature Selection achieved the highest prediction accuracy (0.969), with high sensitivity (0.857) and specificity (1.000).
• Key predictors for structural epilepsy included age at first seizure, cluster seizures, and seizure frequency within various timeframes.

Conclusions:

• Machine learning models, specifically Bayesian Networks and Random Forests, can effectively identify dogs with structural epilepsy.
• These algorithms demonstrate high sensitivity and specificity in predicting canine structural epilepsy.
• The findings offer valuable guidance for veterinarians and pet owners in the clinical decision-making process for seizuring dogs.