Artificial intelligence-assisted smartphone thermal imaging for pedicled facial and head-neck flaps: An exploratory feasibility study using ΔT-based proxy perfusion grading

This pilot study evaluates the feasibility of artificial intelligence (AI)-assisted analysis of smartphone-based thermal imaging (SBTI) for perfusion assessment in pedicled facial, head and neck flaps. Because SBTI provides only false-color, non-radiometric thermal maps, ΔT-based grading relies on subjective human interpretation. We therefore investigate whether AI-derived perfusion probability maps were explored in relation to this manually generated proxy reference.

Twenty-six patients undergoing reconstruction with pedicled skin flaps were examined with SBTI at four perioperative timepoints. ΔT-values were visually extracted from the FLIR color scale across predefined flap zones and translated into a five-grade perfusion score. A fine-tuned ResNet-50 model generated image-level probabilistic classification outputs regarding thermal similarity to patterns labeled as perfused versus non-perfused and uncertainty metrics. Analyses focused on T3 (end of surgery).

Thermal patterns varied substantially by flap type, thickness, and anatomical region. Random-pattern glabellar, nasal, and alar flaps showed physiologically lower surface temperatures and correspondingly reduced ΔT- and AI-derived prediction probabilities despite uneventful healing. No statistically significant correlation was observed between AI-derived perfusion probabilities and manually derived ΔT grades.

AI-assisted SBTI is technically feasible but did not demonstrate meaningful quantitative agreement with manually derived ΔT-based grading in this small exploratory cohort. The findings primarily highlight the methodological limitations of false-color thermal interpretation and the need for radiometric imaging, validated perfusion references, and larger anatomically stratified datasets.