Facial pain intensity estimation for ICU patient with partial occlusion coming from treatment

Pain is a vital warning of threats to health, and facial expressions are often credible indicators of pain intensity. Clinically, intensive care unit (ICU) patients are those most in need of an automatic facial expression pain intensity estimation system. There are numerous ICU patients with nasogastric tubes, which are fixed on the faces with tape and padded with gauze to prevent constant scratching against the skin. Those stuff prevents the effective use of existing automatic facial pain intensity estimation systems, which are designed for unobstructed faces. To estimate the facial pain intensity for such patients, we proposed a method that combines AAM (Active Appearance Model), CNN (Convolutional Neural Network) and ELM (Extreme Learning Machine). AAM was used to generate masks to remove the obstructed facial areas, and followed by inpainting to reduce the interference caused by abrupt brightness change at the mask border. CNN was employed for feature extraction and the pain intensity was regressed via ELM. Experimental results reveal that the proposed method outperforms the previous method in terms of mean squared error and Pearson correlation coefficient.