Premium
Poster — Wed Eve—09: Quest for a “Gold Standard” for Breast Density Evaluation
Author(s) -
Uyaniker B,
Rajapakshe R,
Gordon P,
Silver S
Publication year - 2009
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.3244113
Subject(s) - gold standard (test) , thresholding , breast density , kappa , cohen's kappa , population , statistics , standard error , mathematics , sample (material) , artificial intelligence , standard deviation , computer science , pattern recognition (psychology) , breast cancer , medicine , image (mathematics) , mammography , physics , geometry , environmental health , cancer , thermodynamics
We have examined the breast density evaluation process using commonly employed methods, visual BIRADS estimate, standard thresholding method (Cumulus) and a newly developed automatic assessment algorithm (Bden) in order to obtain clues for a “gold standard” for breast density estimation. According to these results the experts were in exact agreement for 23 out of 36 images, corresponding to the 63.9 % of the total sample and kappa statistics with κ = 0.48 indicates the proportion of the chance agreement (expected) is as 30.9 %. Agreement between the two experts found to be less than 70%. It was always possible to include a given image one class lower or higher. Furthermore, systematical shifts in the BIRADS category estimates between the readers have been noticed. The results obtained, although within a limited sample and from only one patient population, indicate that it is difficult to obtain a gold standard using the visual method as judgment plays an important role. Furthermore, there exist systematical shifts in the BIRADS category estimates between the observers. As a result, assessments given by different observers or obtained at a later time are not easily comparable. Hence, there may be even no “gold standard” unless a new scale of breast density using an automated method is defined.