A Novel Deep Learning Model for Pain Intensity Evaluation

Abstract

Pain assessment is a critical component of healthcare, influencing effective pain
management, individualized care, identification of underlying issues, and patient
satisfaction. However, the subjectivity and limitations of self-reported assessments have
led to disparities in pain evaluation, particularly in vulnerable populations such as
children, the elderly, individuals with cognitive impairments, and those with mental
health conditions. Recent advances in technology and artificial intelligence (AI) have
paved the way for innovative solutions in pain intensity evaluation.This paper presents a
novel deep learning model to automatically classify pain intensity levels and compares
them with six state-of-the-art deep learning classification models - ResNet-50, VGG-19,
EfficientNet, DenseNets, Inception, and Xception- using the UNBC-McMaster Shoulder
Pain Expression Archive Database for training. Transfer learning is employed to optimize
model efficiency and minimize the need for extensive labeled data. Model evaluations
are conducted based on accuracy, precision, recall, and F1 score. The proposed model,
ZNet, showed superior performance of 95.4%, 64.4% and 63.4%, 63.7% for accuracy,
precision, recall and F1-score respectively. Furthermore, this study addresses the
challenge of accurately evaluating pain intensity in patients who cannot communicate
verbally or face language barriers. By harnessing AI technology and facial expression
analysis methods, we aim to provide an objective, reliable, and precise pain assessment
methodology. Automated artificial based solutions enhance the reliability of pain
evaluations, and holds promise for improving decision-making in pain management and
treatment processes, ultimately enhancing patients' quality of life.