Automated pain detection using STA-LSTM on facial landmarks

This research introduces an advanced deep learning system that seamlessly integrates facial landmark extraction, 3D normalization, and a Spatial-Temporal Attention Long Short-Term Memory (STA-LSTM) model for accurate pain intensity level estimation. The video dataset, sourced from hospitals, underwent meticulous categorization and deidentification processes. To overcome dataset imbalances, a thoughtful sampling technique was employed. The STA-LSTM model, trained with optimized parameters, demonstrated an impressive 98% accuracy during training, and validation yielded a substantial accuracy of 92.2%. This proposed approach establishes a comprehensive framework for automated pain assessment, showcasing potential enhancements in patient care within clinical settings. Moreover, the system seamlessly integrates into a user-friendly graphical user interface (GUI).