A enigmatic species, domestic felines. They excel in concealing their emotions and motives, unlike canines, possibly due to their solitary hunting lineage. Detecting indications of discomfort in a cat’s facial expressions and behaviors can pose a challenge for cat guardians and professionals because of this inherent inscrutability. However, emerging artificial intelligence initiatives may ultimately penetrate this facade.
An AI research team and veterinarians developed and tested two machine-learning algorithms aimed at discerning pain in cats under medical care based on their facial cues. A recent publication in Scientific Reports highlighted the efficacy of these integrated systems, achieving up to 77 percent accuracy. This breakthrough signifies the potential for innovative clinical tools in feline healthcare.
Anna Zamansky, a machine learning expert at the University of Haifa in Israel and co-senior author of the study, envisions developing applications that can swiftly identify pain for both professionals and cat owners. Zamansky asserts that despite previous attempts by AI designers to unravel the enigma of animal emotions (such as a 2021 game called Tably), this study marks the first peer-reviewed scientific exploration in this domain.
Currently, professionals rely on intricate assessments like the Glasgow Composite Determine Pain Scale to evaluate feline distress, which involves meticulously analyzing facial expressions and behaviors. Although these scales are statistically validated, they are time-intensive and subject to subjective interpretation by veterinarians, as noted by clinical psychologist Stephane Bleuer from Tel Aviv, who was not affiliated with the research.
Zamansky expressed optimism about the capabilities of their technology, stating, “We believe the system will outperform existing methods.” Due to its sensitivity to subtle sensory cues, the AI system can discern nuances beyond human perception.
To train their novel concept, researchers collected data at the University of Veterinary Medicine Hannover’s animal clinic in Germany, capturing images of 84 cats of various breeds and ages with diverse medical histories. These images, along with the cats’ known medical conditions and corresponding pain levels assessed using the Glasgow scale, were utilized to enhance the team’s AI models and evaluate their efficacy. The researchers emphasized that none of the cats experienced any distress during the study.
The team developed two machine learning algorithms that could detect pain solely from cat photographs. One algorithm focused on 48 facial “landmarks,” including the ears, eyes, and mouth, to gauge facial muscle contraction—a common indicator of pain. The second algorithm utilized deep learning techniques to analyze the entire face for muscle tension and patterns associated with discomfort.
While the landmark-based AI technique boasted a 77 percent accuracy rate in identifying pain, the deep learning approach achieved a slightly lower accuracy of 65 percent. The researchers attributed this variance to the data-intensive nature of deep learning systems, especially considering the relatively modest image dataset utilized in the study.
According to Sebastian Meller, a veterinarian at the University of Veterinary Medicine and co-author of the study, the research unveiled that the cat’s mouth, rather than the eyes or ears, played a pivotal role in accurate pain recognition. This unexpected discovery underscores the AI’s capacity to unveil insights previously overlooked by human observers.
However, Dennis Küster, a sentiment science expert not involved in the study, emphasized the importance of distinguishing between facial expressions and emotions. AI excels at identifying visual patterns but may not always grasp the underlying emotional context, as evidenced by human studies. Küster illustrated this point with the example of smiling, which may not always signify genuine happiness.
Despite these nuances, AI holds promise in emotion recognition, particularly in scenarios where verbal communication is absent, as highlighted by Brittany Florkiewicz, a psychology professor at Lyon College. Florkiewicz stressed the significance of feeding AI systems diverse, supervised datasets to enhance their efficacy in interpreting nuanced information accurately.
Florkiewicz’s research on animals’ facial expressions, revealing 276 distinct expressions, aligns with Zamansky’s goal of delving deeper into feline behavior beyond pain assessment. Zamansky aims to extend her research to encompass other species like dogs, exploring the potential of automated systems in assessing animal well-being based on full-body cues.
A user-friendly pain detection app could revolutionize feline care by enabling prompt identification of issues that might have otherwise gone unnoticed, enhancing both feline and human well-being, according to Bleuer. The collaborative efforts of the international team, comprising scientists, engineers, veterinarians, and AI experts, exemplify the interdisciplinary approach required to bridge interspecies communication barriers and advance animal welfare.
Zamansky reflected on the transformative journey of the research team, transcending language barriers and disciplinary boundaries, culminating in a collective effort that promises to benefit not only cats and their caregivers but a myriad of animal species. This endeavor has even kindled Zamansky’s newfound appreciation for cats, transitioning her from a dedicated “dog person” to someone considering feline companionship.
