Lightweight Multimodal Algorithm for Automated Pain Detection

Pain assessment is crucial in medicine, as it indicates patient health and guides diagnosis and treatment. Traditional methods have significant limitations, especially for children, unconscious patients, and those in intensive care. Therefore, this study proposes a multimodal automated pain assessment algorithm that analyses facial expressions, vocal sounds, and variations in facial temperature. The algorithm integrates multimodal data through fusion at both the feature level and the decision level. First, it combines data from facial temperature and expressions to create a bimodal model. Then, it trains an audio model separately. The algorithm unifies these models to generate decision values that are combined into a new feature matrix for prediction. Therefore this lightweight multimodal pain assessment algorithm reduces complexity and cost while ensuring accuracy, offering broader and more effective pain management for the general public.