A baby's cry after surgery is currently assessed by a nurse glancing at a chart, but that snapshot can miss pain that comes and goes unpredictably. This matters because existing pain scoring for infants relies on intermittent, subjective checks of behaviour and vital signs—leading to unreliable identification and under-treatment of pain. Pain in early life is linked to long-term adverse neurodevelopmental outcomes, and parents find it deeply concerning. The researcher will use machine learning to analyse continuous video, audio, and vital signs data from infants after surgery, aiming to automate pain assessment. They will also explore EEG data for features associated with ongoing pain, and investigate how parents’ experiences could shape future clinical tools. If successful, this work could generate continuous, objective pain profiles, allowing clinicians to time pain-relieving interventions more precisely. The immediate impact would be on hospital wards—reducing “missed pain” in infants who cannot speak for themselves. Beyond that, the multimodal approach could inform future automated monitoring systems for other vulnerable patients, though this remains a fundamental science project focused on developing and validating the method itself.
View original technical description
Established infant pain scores primarily rely on intermittent, subjective assessments of behavioural and physiological activity (1). This leads to unreliable identification and therefore treatment of pain. This is a particular problem for infants at risk of ongoing pain, for example after surgery, since the timing and intensity of pain can be unpredictable (1). Pain in early life is associated with long-term adverse neurodevelopmental outcomes (2), and is a cause of great concern for parents (3). The development of continuous, objective pain assessment methods could enable clinicians to optimise timing of pain-relieving interventions. Highly valuable temporal profiles of infant pain could be generated and the risk of "missed pain" in infants reduced. This could substantially improve infant well-being. Using state-of-the-art machine-learning methods with continuous video and vital signs data, it may be possible to develop a continuous infant pain assessment tool. To date, studies have mainly explored video analytics for acute procedural pain detection in infants (4,5). One study explored infant post-surgical pain, but included only nine patients videoed for three hours, and did not include vital signs (6). The importance of using multiple modalities for infant pain assessment is widely accepted (6,7). In my proposed DPhil, I will use machine learning to automate the clinical assessment of post-surgical infant pain using con1nuous video, audio, and vital signs data. I will analyse EEG data to discover features associated with ongoing pain. I will explore how parental experiences could inform future clinical development of these technologies. Project 1 - Making subjective pain scoring more objective. Aim: Automate the clinical assessment of post-surgical pain Project 2 - Making intermittent pain scoring continuous. Aim: Describe the frequency and intensity of post-surgical pain Project 3 - Exploring EEG features associated with post-operative pain. Aim: Explore EEG features associated with "symptomatic" and "no" pain scores. Project 4 - Exploring parents' experiences. Aim (Part A): Compare automated and staff pain assessments to parents' assessments. Aim (Part B): Explore parents' experiences of pain assessment. References 1. E.M. Boyle et al.,Acta Paediatr.107, 63-67 (2018). 2. B.O. Valeri et al., Clin J Pain. 31, 355-362(2015). 3. L.S. Franck et al.,Arch Dis Child-Fetal. 89, 71F-75(2004). 4. J. Yan et al., IEEE T Affect Comput. 14, 245-254 (2023). 5. J. Egedeet al., Int Conf Affect. 2019, 1-7 (2019). 6. M.S. Salekin et al., Comput Biol Med. 129, 104150(2021). 7. M. van der Vaart et al., Paediatr Neonatal Pain. 1, 21-30(2019).
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know