CNS*2026 Halifax

Introduction
Cerebral palsy (CP) is the most common childhood motor disability, with a prevalence of 1.6 per 1000 births worldwide [1]. A common symptom of CP is chronic pain, with 76% of children experiencing pain, and 33% experiencing chronic pain [2]. Existing pain assessment tools rely on self- or proxy-reporting, limiting their utility for children with communication or cognitive impairments [3]. Electroencephalography (EEG) offers a non-invasive and objective alternative by identifying neural biomarkers associated with pain [4], [5], [6]. This study aims to develop and evaluate machine learning models for classifying pain intensity in children with CP using EEG data.


Methods
Ten children with cerebral palsy and chronic pain, along with ten age-matched healthy controls, will undergo EEG recording during a hamstring stretching protocol administered by a research physiotherapist [6]. Measurements will occur in three conditions including rest, non-painful, and painful stretching. The intensity of pain will be continuously monitored and measured by either the Visual Analog Scale (VAS) for verbal participants, or the Faces Pain Scale – Revised (FPS-R) for non-verbal participants. The EEG data will then be processed, and power spectral density will be computed across all frequency bands. A support vector machine, and two deep-learning models will be evaluated on their ability to accurately classify pain EEG signals.

Results
We hypothesize that children with CP will exhibit increased theta and alpha power in the somatosensory and frontal cortices during painful stretching, in accordance with previous literature, while controls will exhibit the opposite pattern [7]. Chronic pain may also alter ERP components such as N100 and P300, reflecting abnormal cognitive processing, with greater modulations in individuals with CP due to sensorimotor impairments [7]. For classification, SVM is expected to provide strong baseline performance, while deep learning models are anticipated to outperform SVM across accuracy, sensitivity, specificity, and F1 score by capturing more nuanced frequency-specific pain patterns.


Discussion
This research holds important clinical relevance, particularly for children with CP who have been historically underrepresented in pain assessment research due to communication challenges and the subjective nature of traditional pain assessment methods. By developing objective, EEG-based biomarkers for pain detection and intensity classification, this research will fill a critical gap in pediatric pain management. Accurate identification of pain could lead to more personalized and effective treatment strategies. Ultimately, this research may inform the development of tools that could be integrated into clinical settings to support clinicians in making faster data-driven decisions about pain interventions, thus improving quality of life.


References

1. Rosenbaum, P., et al. (2007). Developmental Medicine and Child Neurology. Supplement, 109, 8–14.
2. Harvey, A., et al. (2024). BMC Medicine, 22(1), 238. https://doi.org/10.1186/s12916-024-03458-0
3. Shauna Kingsnorth, et al. (2018). https://hollandbloorview.ca/research-education/knowledge-translation-products/chronic-pain-assessment-toolbox-children
4. Rockholt, M. M., et al. (2023). Frontiers in Neuroscience, 17, 1186418. https://doi.org/10.3389/fnins.2023.1186418
   
5. Chmiel, J., et al. (2025). Journal of Clinical Medicine, 14(16), 5902. https://doi.org/10.3390/jcm14165902
6. Sabater-Gárriz, Á., et al. (2024). Research in Developmental Disabilities, 150, 104760. https://doi.org/10.1016/j.ridd.2024.104760
7. dos Santos Pinheiro, E. S., et al. (2016). http://hdl.handle.net/20.500.12105/20252
   

Acknowledgement
N/A