According to a study presented at the ANESTHESIOLOGY® 2023 annual meeting, an automated pain detection system based on artificial intelligence (AI) has promise as an impartial technique to detect pain in patients before, during, and after surgery.
Image Credit: NicoElNino/Shutterstock.com
To assess pain, subjective methods such as the Visual Analog Scale (VAS)—in which patients rate their own pain—and the Critical-Care Pain Observation Tool (CPOT—in which health care professionals rate the patient’s pain based on facial expression, body movement, and muscle tension, are currently used. The automated pain detection system employs two types of AI: computer vision (which gives the computer “eyes”) and deep learning to understand images to measure patients’ discomfort.
Traditional pain assessment tools can be influenced by racial and cultural biases, potentially resulting in poor pain management and worse health outcomes. Further, there is a gap in perioperative care due to the absence of continuous observable methods for pain detection. Our proof-of-concept AI model could help improve patient care through real-time, unbiased pain detection.
Timothy Heintz, Study Lead Author, University of California San Diego
Early pain diagnosis and treatment have been found to reduce hospital visits and avoid long-term health issues such as chronic pain, anxiety, and depression.
Researchers fed the AI model 143,293 facial images from 115 pain episodes and 159 non-pain episodes in 69 patients undergoing a variety of elective surgical operations ranging from knee and hip replacements to major cardiac surgeries.
The computer was educated by the researchers by showing each raw facial image and informing it whether or not it reflected pain, and it learned to recognize patterns.
The researchers discovered that the computer concentrated on facial emotions and facial muscles in certain parts of the face, including the brows, lips, and nose, using heat maps. It utilized the learned information to predict pain after being shown enough instances. The AI-assisted pain identification system matched CPOT findings 88% of the time and VAS results 66% of the time.
Heintz added, “The VAS is less accurate compared to CPOT because VAS is a subjective measurement that can be more heavily influenced by emotions and behaviors than CPOT might be. However, our models were able to predict VAS to some extent, indicating there are very subtle cues that the AI system can identify that humans cannot.”
If the findings are confirmed, this technology might be another tool for clinicians to employ to improve patient care. Cameras positioned on the walls and ceilings of the surgical recovery room (post-anesthesia care unit), for example, might be used to measure patients’ suffering—even those who are asleep—by collecting 15 images per second.
This would also allow nurses and other health professionals to focus on other aspects of care rather than periodically assessing the patient's discomfort. The researchers intend to continue including more elements into the model, such as movement and sound.
Concerns regarding patient image privacy would need to be addressed, but the system could perhaps integrate additional monitoring elements, such as brain and muscle activity to check unconscious patients, he added.