Search for a command to run...
Accurate pain assessment remains a cornerstone of effective clinical care as it significantly influences diagnosis, treatment planning, and evaluation of therapeutic outcomes. Traditional pain assessment tools such as the Visual Analog Scale (VAS), Numerical Rating Scale (NRS), and Verbal Rating Scale (VRS) rely heavily on the patient's ability to self-report their level of discomfort. However, these conventional approaches are inadequate for patient populations with impaired communication abilities, including individuals with neurological disorders, dementia, or those in postoperative recovery. To overcome these challenges, this study presents a novel, automated pain assessment framework that uses artificial intelligence (AI) and facial expression analysis to objectively quantify pain levels. The proposed system incorporates transfer learning and deep neural network models to improve the accuracy of pain detection using facial cues. Using the UNBC-McMaster Shoulder Pain Expression Archive Database, a widely recognized benchmark in pain research, the model was trained to identify and classify facial expressions associated with different levels of pain. A key innovation of this research is the development of an enhanced multilevel pain scale, which extends the traditional ten-point scale to sixteen different levels, allowing for more precise and granular assessment. Despite the inherent problem of class imbalance within the dataset, the model achieved a commendable classification accuracy of 91%. The results highlight the viability of AI-based tools as reliable, non-invasive alternatives to traditional self-report methods, particularly for non-communicative patients. This advancement promises to improve patient care by supporting clinicians with objective, data-driven pain assessment techniques.