Automated video-based pain recognition in cats using facial landmarks

Area of Science:

• Veterinary Medicine
• Animal Behavior
• Artificial Intelligence

Background:

• Facial expressions in mammals reflect affective states, with pain recognition being a key area of research in non-human animals.
• Automated pain recognition systems aim to overcome the subjectivity and labor-intensive nature of manual facial expression analysis in animals.
• Previous AI approaches for cat pain recognition relied on static images and manual landmark annotation.

Purpose of the Study:

• To develop and evaluate a fully automated, end-to-end artificial intelligence (AI) pipeline for pain recognition in cats using video data.
• To investigate the impact of temporal information from video on pain recognition accuracy compared to static image analysis.
• To define and analyze metrics for dataset deficiencies in animal pain facial expression studies.

Main Methods:

• Development of an end-to-end AI pipeline processing video data for automated pain detection in cats.
• Elimination of manual intervention in image selection and facial landmark annotation.
• Utilizing temporal dynamics from video sequences for enhanced pain recognition.

Main Results:

• The presented AI pipeline achieved over 70% and 66% accuracy on two distinct cat pain datasets.
• The video-based approach outperformed previous automated methods that used single static frames.
• Analysis of dataset deficiencies and their impact on AI performance was conducted.

Conclusions:

• The integration of temporal information from video significantly enhances automated pain recognition in cats.
• The developed end-to-end AI pipeline offers a more efficient and objective method for feline pain assessment.
• Future research should consider the dynamics of facial expressions for more accurate animal pain detection.