Development and Validation of an IMU Sensor-Based Behaviour-Alert Detection Collar for Assistance Dogs: A...

Area of Science:

• Biomedical Engineering
• Canine Behavior Science
• Machine Learning Applications

Background:

• Assistance dogs aid epilepsy management, but alert detection lacks standardization and objective validation.
• Current methods for seizure alert detection by dogs are inconsistent and difficult to quantify.
• Wearable technology offers potential for objective, automated monitoring of canine assistance behaviors.

Purpose of the Study:

• To develop and validate a wearable behavior-alert detection collar for trained assistance dogs.
• To demonstrate the technical feasibility of automated detection for specific canine seizure signaling behaviors.
• To establish a foundation for real-time seizure-alerting systems using canine-assisted healthcare.

Main Methods:

• Development of a wearable collar integrating inertial sensors (accelerometer, gyroscope).
• Implementation of a machine learning pipeline for detecting a trained 'spin' alert behavior.
• Evaluation of four supervised learning models (Random Forest, Logistic Regression, Naïve Bayes, SVM) using Leave-One-DOG-Out cross-validation on data from six dogs.

Main Results:

• The Random Forest model achieved the highest performance with an F1-score of 0.65 and 92% accuracy.
• The system demonstrated technical feasibility for automated detection of trained canine alert behaviors.
• A limited dataset of 135 labeled spin alerts was collected, reflecting real-world training constraints.

Conclusions:

• The developed wearable collar system is a novel step in combining canine behavior with wearable technology for medical applications.
• This proof-of-concept study validates the technical feasibility of automated seizure alert detection in assistance dogs.
• The system provides a foundation for future development of scalable, animal-assisted healthcare innovations and Internet of Medical Things solutions.