Automatic video-based discomfort detection for premature infants


Preterm birth is a term used when a baby is born too early, before 37 weeks of pregnancy have been completed. Based on statistics for 184 countries, the global average preterm birth rate in 2010 was 11.1%, giving a worldwide total of 14.9 million infants. Neurobiological vulnerability to pain in preterm infants is well established, due to their lower pain threshold and immature systems for maintaining homeostasis. Frequent pain or discomfort in preterm infants, who undergo prolonged hospitalization at a time of physiological immaturity and rapid brain development, can cause complications, such as delay in cognitive and motor development. Cumulative pain-related discomfort may also contribute to abnormal brain development, which can yield long-term adverse neurodevelopmental outcomes. In this regard, continuous monitoring of preterm infants for discomfort or pain is necessary. Premature infants receive special care in the Neonatal Intensive Care Unit (NICU), where their vital signs are continuously monitored. However, there is currently no system for monitoring and detecting their pain or discomfort. Monitoring by healthcare professionals is of high costs, time-consuming and subjective in assessment. Because of these reasons, infants are only observed during short intervals a few times a day, which possibly leaves many discomfort moments unnoticed. Therefore, an automatic video-based discomfort detection system is needed, which enables discomfort detection by analyzing facial expressions.
The facial expressions are important factors since the Primal Face of Pain, shown in Figure 1, is a universal facial expression, associated with pain, which is hardwired and present at birth. It is important that the analysis of videos is inherently an unobtrusive action for infants.


Our recent publications have showed the feasibility of video-based discomfort detection for infants. There are still tasks for the future work to improve the current system. This involves a number of research activities, which are listed below.

  • Discomfort detection using motion information
  • Audio analysis for discomfort detection
  • Data augmentation to increase training data size for deep learning
  • Evaluation of the developed face tracking method and comparison with existing methods
  • Evaluation of discomfort detection performance of the developed model and comparison with other state-of-the-art methods
This project will be mainly executed internally at the TU/e, but also partly at Veldhoven MMC hospital.
machine learning, video content analysis, video, healthcare/perinatology
Contact person: