Automatic early detection of induced colic in horses using accelerometer devices.

Area of Science:

• Equine health
• Veterinary diagnostics
• Artificial intelligence in animal health

Background:

• Early recognition of colic signs in horses is crucial for timely veterinary care.
• Subjective behavioral observation by owners has limitations, including time constraints and difficulty detecting subtle signs.
• Advancements in wearable technology and AI offer potential for automated colic detection systems.

Purpose of the Study:

• To develop and validate a software algorithm for the early detection of colic in horses.
• To assess the algorithm's capability in differentiating between varying levels of colic-induced pain.

Main Methods:

• Induction of transient colic in eight mares using prostaglandin.
• Collection of accelerometric and video data alongside veterinary pain scoring (none, level 1, level 2).
• Development of AI models using behavioral and activity data to detect colic and assess pain severity, validated against veterinary assessments.

Main Results:

• The algorithm achieved 91.2% accuracy in detecting colic and 93.8% accuracy in differentiating between pain levels 1 and 2.
• The system accurately classified 10 distinct pain-related behaviors and distinguished them from normal behavior.
• High accuracy was demonstrated in classifying behaviors and activity indices for colic detection.

Conclusions:

• The developed system offers a unique and innovative approach to early colic detection in horses.
• The algorithm can effectively distinguish between different severities of colic.
• While the dataset had limitations in severe pain cases, the core algorithm shows strong potential for early colic sign identification.