Machine learning and explainable AI for pain assessment and clinical decision support
Melpo's research investigated three complementary approaches to improving how pain is managed, measured, and acted upon in clinical care: Virtual Reality (VR), Machine Learning (ML), and Explainable Artificial Intelligence (XAI).
Her systematic reviews showed that VR can meaningfully reduce acute procedural pain and improve emotional well-being in cancer patients, with positive effects reported in the large majority of studies reviewed, though the evidence base remains methodologically inconsistent and needs more standardized, long-term clinical validation.
Her machine learning work demonstrated that pain intensity can be estimated automatically from facial video and images, using deep learning models trained on a public pain-intensity benchmark dataset. This offers a path toward objective pain assessment for patients who cannot reliably self-report pain, such as people with advanced dementia though her results also show that today's models still face real limits in reliability and consistency that would need to be resolved before clinical use.
Her final contribution was an explainable AI system, called Cognica, that supports clinicians managing a rare, painful genetic condition called schwannomatosis. Rather than acting as a black-box tool, Cognica explains its reasoning in terms a clinician can directly verify against established clinical guidelines. In an evaluation with clinical experts, this explainability was shown to significantly increase clinicians' trust and understanding supporting the broader conclusion of the thesis: that in healthcare AI, being able to explain a recommendation is not a luxury, but a core requirement for a tool to be safely adopted.