Comparison between AI and human expert performance in acute pain assessment in sheep

Area of Science:

• Veterinary Medicine
• Artificial Intelligence
• Animal Welfare

Background:

• Accurate animal pain recognition is crucial for welfare and effective treatment.
• Current methods rely on subjective human interpretation, which can be inconsistent.
• Developing objective pain assessment tools is a significant challenge in veterinary science.

Purpose of the Study:

• To investigate if Artificial Intelligence (AI) can surpass human veterinary experts in identifying pain in sheep.
• To compare AI performance against established pain scoring scales used by experts.
• To explore the potential of AI in enhancing the objectivity and accuracy of animal pain recognition.

Main Methods:

• A dataset of 48 sheep videos was collected, recorded before and after surgery to capture pain states.
• Veterinary experts assessed pain using the Sheep Facial Expression Pain Scale (SFPES) and the Unesp-Botucatu Composite Behavioural Scale (USAPS).
• An AI pipeline utilizing a CLIP encoder was developed and trained on facial images for pain recognition.

Main Results:

• The AI pipeline significantly outperformed human experts in facial pain scoring (AUC difference = 0.115, p < 0.001).
• AI performance in behavioral scoring (USAPS) was comparable to human experts, with a non-significant improvement (AUC difference = 0.027, p = 0.163).
• The AI demonstrated superior ability in detecting pain cues from visual facial information.

Conclusions:

• AI models can outperform human experts in specific aspects of animal pain recognition, such as facial expression analysis in sheep.
• AI shows potential as a valuable tool to aid veterinarians in objective pain assessment, complementing existing methods.
• Further research and validation are warranted to integrate AI into routine clinical practice for improved animal welfare.