Pain analysis in newborns has become a relevant study subject over the last few decades, given the inability to objectively identify the source and intensity of the pain in newborn babies. Over the last few years, several methods for pain detection and evaluation were able to classify pain levels using facial expressions from newborn babies, through statistical models, machine learning and deep learning. Considering this context, health professionals are increasingly more interested in having computerized tools at their disposal. These tools would not only be able to accurately rank the newborn’s potential pain level, but also identify the facial regions of greatest relevance for a particular pain phenomenon. This dissertation’s main objective is to develop a computer framework capable of recognizing and interpreting patterns in facial expressions for an automated evaluation of pain levels on term babies. Specifically, this dissertation focuses on the investigation, implementation and integration of a series of techniques, including image detection and segmentation, spacial normalization and, ultimately, the classification of facial expressions based on information obtained through statistical data mining. Finally, the framework developed here, evaluated with an accuracy (upper limit) of approximately 96% for the COPE base and 77% for the UNIFESP base, reveal that it is possible to not only rank pain levels statistically through images of facial expressions, but also to identify key facial regions for certain pain phenomena, therefore assisting in creating more general and accurate pediatric pain scales
A Robot Has a Mind of Its Own Because We Intuitively Share It
