Semi-automated erythema quantification in infantile haemangioma receiving propranolol: a pilot feasibility study.

Abstract

Background: Infantile haemangioma (IH) is the most common vascular tumour of infancy. Propranolol is first-line treatment, but response monitoring relies on subjective clinical assessment. Semi-automated photographic erythema quantification may offer a reproducible, low-cost adjunct monitoring tool.

Objectives: To develop a semi-automated Python-based erythema quantification tool for serial clinical photographs of IH, evaluate its feasibility and intra-rater reliability, and characterise the relationship between colour-based erythema change and clinician-assessed composite response.

Methods: Retrospective analysis of a prospective photographic registry at a tertiary paediatric surgery centre. Forty-nine infants with IH treated with propranolol were included. Serial clinical photographs (n = 251) were analysed using a custom Python tool. The erythema index (EI = [R - G] / [R + G + B]), a formula validated for quantification of cutaneous erythema from digital photographs, was calculated from manually selected lesion regions of interest (ROI). Intra-rater reliability was assessed by re-rating 22 photographs from five patients one week apart. Absolute EI change was correlated with prospectively recorded, blinded clinician response scores (0-10).

Results: Of 251 photographs, 242 (96.4%) were analysable (mean 4.9 visits per patient, range 2-11). Intra-rater reliability was high: ICC(A,1) = 0.990(95% CI 0.97-1.00); Bland-Altman mean difference - 0.005 EI units (95% limits of agreement - 0.023 to + 0.013). Mean baseline EI was 0.122 ± 0.058 and mean final EI was 0.115 ± 0.053. Absolute EI change did not correlate with clinician response scores (Pearson r = - 0.084, 95% CI - 0.357 to 0.202, p = 0.566; Spearman r = - 0.087, p = 0.554), consistent across three sensitivity analyses.

Conclusions: Semi-automated erythema quantification from clinical photographs of IH is feasible with high intra-rater reliability. These pilot findings should be interpreted as hypothesis-generating; the study is underpowered for definitive correlation testing, and prospective replication in an adequately powered cohort is required. In this pilot cohort, colour-based EI change alone was insufficient to track the clinician's composite response assessment under real-world uncalibrated photography conditions, consistent with the partial-biomarker nature of colour-only analysis. Prospective studies with standardised photography incorporating colorimetric calibration and composite colour-plus-area scoring are warranted.