Objective Pain Assessment with PPG: Measuring Pain from Photoplethysmography Signals

Machine Learning Approaches to PPG-Based Pain Assessment

Feature Engineering for Pain Classification

Beyond SPI and ANI, researchers have explored multimodal physiological feature sets for pain classification.

Werner et al. (2014) (doi:10.1109/icpr.2014.784) combined facial video with biomedical signals (EMG, skin conductance, ECG) from the BioVid Heat Pain Database for automatic pain recognition, showing that multimodal fusion improved recognition over single modalities.

Deep Learning

Recent work has applied deep learning directly to raw physiological signals, bypassing manual feature engineering.

Thiam et al. (2019) designed deep learning architectures on raw physiological signals (EDA, ECG, EMG) from the BioVid Heat Pain Database. Their EDA-only model and deep-fusion models reached average accuracies of 84.57% and 84.40% for baseline vs. pain-tolerance classification under leave-one-subject-out validation, outperforming previous hand-crafted-feature methods. Notably, EDA rather than cardiac signals carried most of the discriminative information.

Lopez-Martinez and Picard (2018) (doi:10.1109/embc.2018.8513575) used recurrent neural networks to estimate continuous pain intensity from autonomic signals including skin conductance and heart rate variability.

Multi-Modal Pain Assessment

PPG alone captures only the autonomic component of pain. Combining PPG with other modalities improves pain assessment:

• PPG + EDA (electrodermal activity): EDA captures sympathetic cholinergic sudomotor responses that are independent of the sympathetic adrenergic responses captured by PPG. Combining both modalities can improve pain classification performance.
• PPG + EMG (electromyography): Facial EMG captures pain-related muscle activation (corrugator, zygomaticus) that is mechanistically distinct from autonomic responses.
• PPG + facial video: Computer vision-based facial action unit detection combined with PPG provides complementary behavioral and physiological pain information.

Clinical Applications and Evidence

Intraoperative Nociception Monitoring

The primary clinical application of PPG-based pain assessment is guiding analgesic administration during general anesthesia. The clinical goal is maintaining adequate analgesia without overdosing, which causes hypotension, delayed recovery, and respiratory depression.

Multiple randomized controlled trials have evaluated nociception index-guided analgesia management. Studies generally report that nociception-guided approaches can reduce intraoperative opioid consumption and hemodynamic instability while maintaining adequate analgesia, though the magnitude of benefit varies across different surgical contexts and patient populations.

ICU Pain Assessment

Critically ill patients in the ICU are frequently unable to self-report pain due to sedation, mechanical ventilation, or cognitive impairment. Current practice relies on behavioral pain scales (BPS, CPOT) that require trained nurses and provide only intermittent assessment.

Continuous PPG-based nociception monitoring could supplement behavioral assessment in the ICU. Ben-Israel et al. (2013) (doi:10.1007/s10877-013-9487-9) developed a multi-parameter nociception level index in 25 surgical patients under general anesthesia. Validation of such indices in ICU populations remains limited.

Neonatal Pain Assessment

Neonates, particularly premature infants, experience numerous painful procedures during NICU stay (heel lances, IV insertions, intubation) and may be inadequately treated due to difficulty assessing their pain. PPG-based indices offer continuous, objective supplementation of behavioral assessment.

Cremillieux et al. (2018) (doi:10.1002/ejp.1191) evaluated the NIPE index in 29 premature neonates during heel lance but found that NIPE did not correlate reliably with behavioral pain scales and was judged unreliable for acute pain assessment in this population.

Chronic Pain Monitoring

PPG-based pain assessment is least validated for chronic pain, where the relationship between nociception and autonomic activation is weakest. Chronic pain patients often show:

• Autonomic habituation (blunted autonomic responses to ongoing pain)
• Central sensitization (pain without peripheral nociceptive input)
• Autonomic dysfunction (baseline autonomic abnormalities confounding interpretation)
• Medication effects (opioids, gabapentinoids, antidepressants alter autonomic function)

Limited evidence suggests that PPG-derived HRV may serve as a biomarker for pain-related autonomic burden in chronic pain populations. A meta-analysis by Tracy et al. (2016) found a consistent moderate-to-large reduction in high-frequency HRV across chronic pain conditions (heavily influenced by fibromyalgia studies), though this likely reflects general autonomic dysregulation rather than pain intensity per se.

For information on autonomic assessment via PPG, see our article on PPG autonomic function testing.

Limitations and Ethical Considerations

Fundamental Measurement Limitations

PPG-based pain indices face several irreducible limitations:

Specificity: Autonomic responses to pain are not distinguishable from autonomic responses to anxiety, stress, fever, hypovolemia, bladder distension, or many other stimuli. Any factor activating the sympathetic nervous system will alter SPI, ANI, and similar indices regardless of whether pain is present.

Sensitivity to medications: Beta-blockers, alpha-agonists (clonidine, dexmedetomidine), anticholinergics (atropine, glycopyrrolate), and vasopressors directly affect the cardiovascular parameters that PPG-based pain indices measure, independently of nociception.

Individual variability: Pain sensitivity, autonomic reactivity, and analgesic requirements vary enormously between individuals. Population-level calibrations provide only rough estimates for any individual patient.

Chronic pain insensitivity: PPG indices primarily capture acute nociceptive responses and have limited sensitivity to chronic pain states where autonomic habituation has occurred.

Ethical Considerations

The pursuit of objective pain measurement raises important ethical questions. Pain is fundamentally a subjective experience, and patients' self-reports should remain the primary basis for pain management decisions when available. PPG-based indices should supplement, not replace, patient communication.

There is a risk that objective pain indices could be used to discredit patient pain reports (if the index reads low, is the patient's complaint dismissed?). Clinical guidelines should ensure that pain indices are used to detect under-reported pain (in non-communicative patients) rather than to override self-reported pain.

Our conditions page provides additional context on how various health conditions affect pain processing and PPG signals.

Future Directions

Multi-Wavelength Pain Assessment

Different PPG wavelengths probe different tissue depths and vascular beds. Multi-wavelength PPG arrays could simultaneously capture superficial cutaneous vasoconstriction and deeper muscular vascular responses to pain, potentially improving specificity. For background on wavelength selection, see our article on green vs. red vs. infrared PPG.

Integration with Other Biomarkers

Combining PPG-derived autonomic pain indices with other objective biomarkers, including pupillometry (pupil dilation reflex), skin conductance (sudomotor response), and EEG-based pain signatures, could create multi-modal pain assessment systems with improved accuracy and specificity.

Personalized Pain Models

Individual calibration using brief standardized pain stimuli (pressure algometry, electrical stimulation) could establish person-specific relationships between nociceptive input and PPG response, enabling personalized pain scoring that accounts for individual autonomic reactivity. This personalized approach, combined with continuous learning from clinical outcomes, represents the most promising path toward clinically useful objective pain assessment through PPG-based health monitoring algorithms.

Conclusion

PPG-based pain assessment provides a window into the autonomic component of nociception, capturing the sympathetic activation and parasympathetic withdrawal that accompany acute noxious stimulation. Established indices like SPI and ANI have demonstrated clinical utility in guiding intraoperative analgesia, reducing opioid consumption, and supplementing pain assessment in non-communicative patients. However, fundamental limitations in specificity, sensitivity to confounders, and individual variability prevent PPG from serving as a standalone pain measurement tool. The technology is most valuable when understood for what it truly measures: not pain itself, but the body's autonomic response to nociceptive input, a physiological correlate that, combined with clinical context and patient communication, can improve pain management decisions.

References

• Huiku, M., et al. (2007). Assessment of surgical stress during general anaesthesia. British Journal of Anaesthesia, 98(4), 447-455. https://doi.org/10.1093/bja/aem004
• Gruenewald, M., et al. (2009). Influence of different remifentanil concentrations on the performance of the surgical stress index to detect a standardized painful stimulus during sevoflurane anaesthesia. British Journal of Anaesthesia, 103(4), 586-593. https://doi.org/10.1093/bja/aep206
• Bergmann, I., et al. (2013). Surgical pleth index-guided remifentanil administration reduces remifentanil and propofol consumption and shortens recovery times in outpatient anaesthesia. British Journal of Anaesthesia, 110(4), 622-628. https://doi.org/10.1093/bja/aes426
• Ledowski, T., et al. (2014). Analgesia Nociception Index (ANI) to predict intraoperative haemodynamic changes: results of a pilot investigation. Acta Anaesthesiologica Scandinavica, 58(1), 74-79. https://doi.org/10.1111/aas.12216
• Jeanne, M., et al. (2012). Variations of the analgesia nociception index during general anaesthesia for laparoscopic abdominal surgery. Journal of Clinical Monitoring and Computing, 26(4), 289-294. https://doi.org/10.1007/s10877-012-9354-0
• Faye, P. M., et al. (2010). Newborn infant pain assessment using heart rate variability analysis. The Clinical Journal of Pain, 26(9), 777-782. https://doi.org/10.1097/AJP.0b013e3181ed1058
• Werner, P., Al-Hamadi, A., Niese, R., et al. (2014). Automatic pain recognition from video and biomedical signals. 22nd International Conference on Pattern Recognition, 4582-4587. https://doi.org/10.1109/ICPR.2014.784
• Thiam, P., et al. (2019). Exploring deep physiological models for nociceptive pain recognition. Sensors, 19(20), 4503. https://doi.org/10.3390/s19204503
• Lopez-Martinez, D., & Picard, R. (2018). Continuous pain intensity estimation from autonomic signals with recurrent neural networks. 40th Annual International Conference of the IEEE EMBC, 5624-5627. https://doi.org/10.1109/EMBC.2018.8513575
• Ben-Israel, N., et al. (2013). Monitoring the nociception level: a multi-parameter approach. Journal of Clinical Monitoring and Computing, 27(6), 659-668. https://doi.org/10.1007/s10877-013-9487-9
• Cremillieux, C., et al. (2018). Objective assessment of induced acute pain in neonatology with the Newborn Infant Parasympathetic Evaluation index. European Journal of Pain, 22(6), 1071-1079. https://doi.org/10.1002/ejp.1191
• Tracy, L. M., et al. (2016). Meta-analytic evidence for decreased heart rate variability in chronic pain implicating parasympathetic nervous system dysregulation. Pain, 157(1), 7-29. https://doi.org/10.1097/j.pain.0000000000000360