Assessment of Pain Intensity Using Deep Learning Models in Non-Communicative Intensive Care Patients.

The study used static facial images, not video, so it did not capture pain expression changing over time. Patients were selected from a single centre and excluded if older than 65, terminally ill, or with facial deformities, paralysis, burns, injuries or obstructed faces, so many ICU patients are not represented. Images were taken under standardised, well-lit, front-facing conditions, which do not reflect routine ICU settings where lighting and equipment can obscure faces. The comparison standard was expert visual rating, not patient self-report, and expert agreement was only slight, with a kappa value of 0.160, especially for moderate pain. Deep sedation or neuromuscular blockade can mask facial expression, so pain may be present without visible facial cues. The authors say future longitudinal studies are needed to test whether AI-driven systems affect patient outcomes, nurse decision-making or resource use. The sample size was determined pragmatically, and the single-centre design limits generalisability.

The authors report nothing to report for funding and declare no conflicts of interest.

Do not conclude this tool can replace bedside pain assessment or detect all pain. It was tested on selected, clearly visible faces under good lighting, and sedation or paralysis can mask facial expression even when pain is present.