Premium
Characteristics of subjective recognition and computer‐aided image analysis of facial erythematous skin diseases: a cornerstone of automated diagnosis
Author(s) -
Choi J.W.,
Kim B.R.,
Lee H.S.,
Youn S.W.
Publication year - 2014
Publication title -
british journal of dermatology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.304
H-Index - 179
eISSN - 1365-2133
pISSN - 0007-0963
DOI - 10.1111/bjd.12769
Subject(s) - rosacea , erythema , dermatology , medicine , decision tree , artificial intelligence , computer science , acne
Summary Background Rosacea and seborrhoeic dermatitis are common diseases that cause facial erythema. They have common features and are frequently misdiagnosed. Objectives To extract characteristic features of erythrotelangiectatic rosacea ( ETR ), papulopustular rosacea ( PPR ) and seborrhoeic dermatitis ( SEB ) through computer‐aided image analysis ( CAIA ) and compare them with subjectively recognized features and to use these findings to construct a decision tree for differential diagnosis. Methods Thirty‐four clinical photos of patients with facial erythema were assessed: 12 patients were classified as showing ETR , 12 as PPR and 10 as SEB . Five dermatologists blinded to the original diagnosis gave their impressions of each photo. The mean, SD and T ‐zone to U ‐zone ( T / U ) ratios of the erythema parameter a* (a* of the L*a*b* colour space) were calculated for each photo using CAIA . These CAIA parameters were compared between impression groups. The most closely related CAIA parameter for each disease was established using the receiver‐operating characteristic curve analysis. A decision tree which predicts the diagnosis from given CAIA parameters was constructed. Results All the photos classified as PPR generated impressions of PPR . However, approximately 30% of the photos classified as ETR generated impressions of SEB and vice versa. PPR was characterized by a large SD of erythema of the cheek, ETR was characterized by a large mean erythema of the U ‐zone, and SEB was characterized by a large T / U ratio of mean erythema. Fifteen additional photos were examined: the decision tree predicted the original diagnosis for 14, but incorrectly predicted one case of ETR as SEB . Conclusions The CAIA result of facial erythema is well correlated with the actual clinical diagnosis. The accuracy of differential diagnosis using a decision tree with CAIA parameters is as good as that of global examination impressions of dermatologists.