How is artificial intelligence helping the diagnosis of pain?

Complex decision-making shrouded in uncertainty is at the core of emergency medical treatment. Visits to the emergency department force doctors and nurses to identify patients with life-threatening conditions from the ones with more common benign diagnoses. Doctors currently use self-assessment models to diagnose pain (Visual Analog Scale VAS). Artificial Intelligence has shown promise in streamlining the diagnosis of pain for medical purposes. We used Google Scholar, ScienceDirect, Springer, and Oxford Academic to perform a literature review. Most research papers got their data and sources from volunteers and established medical databases. Some common databases were the UNBC-McMaster database, the MIntPAIN database, and the BioVid database. These databases collected medical images concerning pain and pain expression. The AI models used a variety of pain scales. There were many Machine Learning models and classifiers that researchers used. The basic models were Convolutional Neural Networks (CNN), Support Vector Machines (SVM), K-Nearest Neighbor, Logistic regression and linear regression models etc. The studies researched the model's capabilities in pain detection and pain intensity estimation. The mean accuracy for the detection of pain among the papers was 85.05%. The mean accuracy for the current pain intensity was 73.90%. The Automatic Coding of the Facial Action Coding System (FACS) is useful for diagnosing pain (7). Developing a pre-trained Machine Learning (ML) model is useful for diagnosing pain. This review confirms that AI/ML technologies can be used to detect pain through facial expressions at a high potential. Artificial intelligence could be a helpful tool in providing objective accurate measurements of pain intensity. It would support doctors and clinicians to make more informed decisions during rush hour emergency moments. An issue of training an Artificial Intelligence model is the need for large amounts of data. The patient's ethical considerations around privacy and algorithm biases must be addressed.