Over one fifth of American adults report experiencing pain on most days or every day. Uncontrolled pain has resulted in increased hospitalization and emergency visits as well as a significant financial burden on patients 1. Measuring pain accurately allows appropriate treatment while preventing overprescription of medications. However, it remains a challenging area. Currently, subjective methods are primarily used to assess pain, including the Visual Analog Scale (VAS), whereby patients rate their own pain, and the Critical-Care Pain Observation Tool (CPOT), whereby health care providers rate a patient’s pain based on their body movement, muscle tension, and facial expression 2. However, some researchers hope that data-driven artificial intelligence (AI) methods will revolutionize our ability to recognize, assess, and treat pain. The clinical goal of using AI for pain monitoring is to develop objective, standardized, and generalizable instruments for assessment across different clinical contexts. To this end, AI-based monitoring methods, which use a diverse range of algorithms, are grouped into behavior-based approaches and neurophysiology-based pain detection methods 3.
Since pain is usually accompanied by spontaneous facial behaviors, a number of different approaches for automatic pain assessment are based on image classification and feature extraction applied to facial expressions. In recent research presented at the Anesthesiology 2023 annual meeting for example, computer vision was developed and trained to feed salient facial information into an AI algorithm in order to determine whether a patient was experiencing pain or not. Researchers provided the AI model over 140,000 facial images from over 110 pain episodes and 150 non-pain episodes across nearly 70 patients who had undergone a wide range of elective surgical procedures, ranging from knee and hip replacements to complex heart surgeries. The researchers taught the computer by showing it raw facial images and telling it whether or not the images represented pain. The computer began to identify patterns, focusing on facial expressions and facial muscles in certain areas of the face, particularly the eyebrows, lips and nose. Remarkably the AI-automated pain monitoring system aligned with CPOT results 88% of the time and with VAS 66% of the time when evaluated on new images.
Neurophysiology-based pain detection attempts to objectively measure pain by monitoring specific markers in the brain or body, including electroencephalography, electrodermal activity, electromyography, and other biosignals. However physiological data as basic and wide-ranging as skin conductance, blood pressure, heart rate, skin temperature and pupillary diameter have also been used to assess pain 4. Most recently, approaches involve multimodal strategies by combining behavior detection with neurophysiological findings. In the future, there remain significant opportunities to leverage this type of multimodal sensing and deep learning to improve accuracy of pain monitoring systems.
One final area to discuss is AI-based computational methods for pain monitoring. Early studies showed that machine learning algorithms such as support vector machine, decision tree, and random forest classifiers could all be used. More recently however, artificial neural networks such as convolutional and recurrent neural network algorithms have been developed, sometimes even in combination with prior strategies 3. As an example, a recent study utilized machine learning, data mining, and natural language processing to improve pain recognition and assessment, analyze self-reported pain data, predict pain, and help both health care practitioners and patients manage chronic pain as effectively as possible 1.
In order to develop the best AI methods for pain monitoring, collaborations among clinicians and computer scientists must be aimed at structuring and processing good quality datasets that can be used across contexts, including both acute to chronic pain conditions 3. In the process, it remains crucial to apply the concepts of explainability and ethics when examining AI applications for pain research and management.
Most reports on AI for pain monitoring to this day remain pilot studies. Therefore, in the future, additional studies are warranted before these approaches are ready for large clinical trials. Future progress in using AI for pain monitoring will improve patient outcomes and healthcare resource utilization.
References
1. Zhang, M. et al. Using artificial intelligence to improve pain assessment and pain management: A scoping review. Journal of the American Medical Informatics Association (2023). doi:10.1093/jamia/ocac231
2. AI pain recognition system could help detect patients’ pain before, during and after surgery | American Society of Anesthesiologists (ASA). Available at: https://www.asahq.org/about-asa/newsroom/news-releases/2023/10/ai-pain-recognition-system. (Accessed: 5th December 2023)
3. Cascella, M. et al. Artificial Intelligence for Automatic Pain Assessment: Research Methods and Perspectives. Pain Research and Management (2023). doi:10.1155/2023/6018736
4. Fernandez Rojas, R., Brown, N., Waddington, G. & Goecke, R. A systematic review of neurophysiological sensing for the assessment of acute pain. npj Digital Medicine (2023). doi:10.1038/s41746-023-00810-1