Objective Detection of Newborn Infant Acute Procedural Pain Using EEG and Machine Learning Algorithms

Area of Science:

• Neuroscience
• Machine Learning
• Neonatal Care

Background:

• Observer-dependent infant pain scales are limited by discontinuous assessments and healthcare professional availability.
• Objective methods are needed to accurately assess acute pain in neonates.
• Electroencephalography (EEG) offers a continuous physiological measure that may capture pain-related neural activity.

Purpose of the Study:

• To investigate the application of agnostic machine learning approaches to neonatal EEG analysis for identifying infant pain responses.
• To develop and validate a machine learning model capable of detecting acute pain in neonates using EEG data.

Main Methods:

• EEG data were recorded from 30 neonates undergoing painful procedures.
• Functional connectivity measures were calculated before and after procedures.
• A gradient boosting machine learning model was trained and validated using leave-one-subject-out cross-validation and an independent test set.

Main Results:

• The optimal gradient boosting model achieved 90% area under the receiver operating characteristic curve, indicating high accuracy in pain detection.
• The model identified 12 key features, primarily related to functional connectivity between specific EEG electrode pairs.
• These features suggest the involvement of brain regions such as the temporal gyrus, opercular cortex, thalamus, and insula in infant pain processing.

Conclusions:

• Machine learning analysis of neonatal EEG, specifically functional connectivity, can objectively detect infant responses to acute pain.
• This approach shows promise in overcoming the limitations of traditional observer-dependent pain scales.
• Future integration of EEG with other physiological and behavioral data could further enhance the assessment of infant pain complexity.