Author(s) :

Aaryan Verma, Anvitha Kakarlapudi, Tamiika Hurst-Darby, PhD.

Volume/Issue :

Abstract :

Optimal chronic pain detection requires objective and timely communication. However, patients are often unable to convey the intensity and length of pain, leading to frequent miscommunication in clinical settings. To address this issue, a machine learning model was developed to detect chronic pain in a timely, less subjective manner. A manually labeled facial image dataset was used for formal model training and held-out testing. Separately, locally collected volunteer images, including a consented image used as an illustrative example, were used only for exploratory demonstration. Three versions were applied, an Action Unit (AU) based model, a Convolutional Neural Network (CNN), and a hybrid model that consists of both AU and CNN features. The hybrid model achieved a testing accuracy of 0.91, outperforming the individual models. Though the small sample size and limited formal chronic pain representation imposed restrictions, this study demonstrates the possibility for AI-based pain detection to be a reliable, on-demand method that can be used alongside clinical examination to improve pain assessment for patients with limited communication.