Computer-Aided Evaluation of Radiologist’s Reproducibility and Subjectivity in Mammographic Density Assessment

Ilijan Tomaš, Zdenka Kotromanović, Nenad Belaj, Damir Margaretić, Zdravko Ivezić, Miroslav Katić, Lada Zibar, Dario Faj, Damir Štimac, Mate Matić

Abstract


Mammographic density is an independent risk of breast cancer. This study has evaluated the radiologists' reproducibility and subjectivity in breast density estimation and in order to decrease the radiologists' subjective errors the computer software was developed. The very good reproducibility existed in  the strong correlation with the first and the second mammogram assessment after three month period for each radiologist (correlation coefficient 0.73-1, p<0.001). The strong correlation was present in the case of all 5 radiologists when compared among themselves and compared with software aided MDEST-Mammographic Density Estimation (correlation coefficient 0.651-0.777, p<0.001). Detected differences in glandular tissue percentage determination occurred in the case of two experienced radiologists, out of 5 (one radiologist with more than 5 year experience and one with more than 10 year experience, p<0.01), but in the case of breast type determination (American College of Radiology-ACR I-IV), the detected difference occurred in one radiologist with the least experience (less than 5 years, p< 0.001). It can be concluded that the estimation of glandular tissue percentage in breast density is rather subjective method, especially if it is expressed with absolute percentage, but the determination of type of breast (ARCI-IV) depends on the radiologist's experience. This study showed that software aided determination of glandular tissue percentage and breast type can be of a great benefit in the case of less experienced radiologists.

Keywords*


mammography, breast density, subjectivity, software

Full Text:

PDF


Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 License.

Current licensing format:

Creative Commons Licence All work is licensed under a  Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.