BG04: Computer-aided quantification of palmar hyperlinearity in atopic dermatitis: a proof-of-concept study

L. Steele,^1,2 B.R. Thomas,^1,2 E.A. O’Toole^1,2 and R.J. Tanaka³

¹Department of Dermatology, The Royal London Hospital, Barts Health NHS Trust; ²Centre for Cell Biology and Cutaneous Research, Blizard Institute, Queen Mary University of London; and ³Department of Bioengineering, Imperial College London, UK

Palmar hyperlinearity is a recognized clinical feature of atopic dermatitis (AD) and has been linked to FLG mutation. The definition of palmar hyperlinearity varies between studies, making direct comparison and clinical interpretation challenging. We aimed to devise a method to quantify computationally palmar hyperlinearity in order to allow rapid, standardized assessment that avoids interobserver variation. Hand images and FLG genotyping were available for 342 participants from the Tower Hamlets Eczema Assessment (THEA) study, which recruited patients of Bangladeshi origin with a dermatologist-confirmed diagnosis of AD. Images were cropped to create a palmar dataset and a thenar eminence dataset. Images were resized to 512 × 256 and Sato edge detection was applied, and greyscale > 0·02 of the resulting images was calculated to generate hyperlinearity scores. Differences in hyperlinearity scores by FLG genotype were compared using one-way anova. We also attempted external validation using images identified from previously published FLG palmar hyperlinearity studies. The FLG mutation status of the 342 participants was 245 (71·6%) wildtype (WT), 87 (25·4%) heterozygous (FLG^+/–) and 10 (2·9%) homozygous/compound heterozygous (FLG^–/–). For the thenar eminence dataset, median hyperlinearity scores for the WT, FLG^+/– and FLG^–/– groups were 11 764 [interquartile range (IQR) 9193–14 671], 16 226 (IQR 12 801–21 206·5) and 26 920 (IQR 18 007–28 680·5), respectively. The difference between groups was statistically significant (P = 0·001). For the palmar dataset, median hyperlinearity scores for the WT, FLG^+/– and FLG^–/– groups were 17 027 (IQR 13 739–21 109), 21 258 (IQR 17 706–25 317·5) and 29 764 (IQR 26 307·25–33 583·75), respectively (P = 0·001). The spectrum of hyperlinearity scores suggested that clinically distinguishing the presence or absence of palmar hyperlinearity can be challenging, with no clear cutoff observed between groups. Six images were available from previously published studies and, when generated scores were mapped onto our data, the results fit with their previous clinical interpretations. Our study is novel in using computational methods to quantify palmar hyperlinearity and also in reporting the association of palmar hyperlinearity and FLG status in a cohort of 342 participants of Bangladeshi origin. The application of our tool in studies investigating palmar hyperlinearity may serve as a metric for comparison between studies, pending wider external validation, and allow for further investigation of hyperlinearity scores for clinical purposes such as the prediction of AD persistence and food allergy risk.