Pain Intensity Assessment with Facial Expressions Using Machine Learning and Deep Learning: A Comprehensive Review

Abstract: In healthcare, a precise evaluation of pain levels is essential for proper diagnosis, treatment planning, and condition monitoring. Due to their reliance on subjective and often inaccurate self-reporting, traditional methods for pain evaluation have their limitations. Machine learning (ML) and deep learning (DL) methods have recently arisen as potential substitutes for objectively assessing the degree of pain. In this study, we survey all the approaches, designs, and data sets used for pain assessment with ML and DL. A variety of methods are discussed in the paper, including feature-based ML algorithms that use synthetic features collected from behavioral and physiological inputs. These algorithms include SVM, RF, and k Nearest Neighbor (k-NN). In contrast, end-to-end DL methods (e.g., CNNs, RNNs, and transformers) make use of raw data derived from multimodal sources (e.g., facial expressions) to improve performance. This paper also delves into hybrid models that combine ML and DL to boost performance. Read this review article to get the lowdown on the most popular model assessment metrics, feature extraction methods, and publically available datasets utilized in this field. We emphasize the difficulties of real-time implementation, dataset biases, and inter-individual heterogeneity. The study outlines recent developments, unanswered questions, and potential future steps to improve the accuracy and utility of pain assessment systems powered by artificial intelligence. Better, more efficient, and scalable methods of measuring pain intensity are the goal of the presented insights, which should help academics and doctors in their pursuits.