Researchers are training artificial intelligence (AI) models to interpret facial expressions as a biomarker of pain, especially for patients who cannot verbally express their pain levels, such as those who are sedated, have dementia, or are nonverbal. An AI algorithm has been developed to analyze images of patients’ faces to detect pain, offering potential for continuous monitoring. This AI system was trained using examples of patients in pain and, using computer vision and deep learning, it learned to recognize pain in other patients.
When compared to the Critical Care Pain Observation Tool (CPOT), the AI’s readings were 88% accurate. The AI focused on areas around the eyebrows, lips, and nose to determine pain. However, its accuracy dropped to 66% when compared with patient-reported measures. The research was presented at the annual meeting of the American Society of Anesthesiologists.