Research presented at the ANESTHESIOLOGY® 2023 annual meeting reveals that an automated artificial intelligence (AI) system for pain recognition is emerging as a reliable method for detecting pain in patients both before and after surgery, as well as during the surgical procedure.
Traditionally, pain assessment has relied on subjective methods like the Visual Analog Scale (VAS) and the Critical-Care Pain Observation Tool (CPOT). However, this new AI-driven pain recognition system leverages computer vision and deep learning to interpret visual cues and evaluate a patient’s pain level.
Early detection and effective pain management have been linked to reduced hospital stays and a lower risk of long-term health issues such as chronic pain, anxiety, and depression.
How the AI System Identifies Pain:
During the study, researchers provided the AI model with 143,293 facial images from 115 instances of pain and 159 non-pain episodes involving 69 patients undergoing various elective surgical procedures.
The AI system learned to identify patterns by analyzing raw facial images, with a particular focus on facial expressions and muscle movements in specific facial regions, including the eyebrows, lips, and nose.
With an adequate number of examples, the AI system was able to make accurate predictions regarding pain levels. It aligned with CPOT results 88% of the time and VAS results 66% of the time.
Dr. Heintz, one of the researchers, pointed out that the VAS is less accurate compared to CPOT due to its subjective nature, which can be influenced by emotions and behaviors. However, the AI system demonstrated the ability to predict VAS scores to some extent, indicating its ability to identify subtle cues that humans may miss.
Future Applications:
If these findings are validated, this technology could become an invaluable tool for healthcare providers to improve patient care.
For example, strategically placed cameras in post-anesthesia care units could continuously assess patients’ pain, including those who are unconscious, by capturing 15 images per second.
This approach could also ease the workload of nurses and healthcare professionals who periodically assess patients’ pain, allowing them to focus on other aspects of patient care.
The research team’s future plans include expanding the model by incorporating additional variables such as patient movement and sound. Addressing privacy concerns will be essential to protect patient images. In the long run, the system could potentially include other monitoring features, such as assessing the brain and muscle activity of unconscious patients.
