Purpose
External validation of a deep learning model for breast density classification based on convolutional neural networks.
Methods and materials
Breast density refers to the relative amount of epithelial and stromal tissue elements (radiopaque) compared with the amount of fatty elements (radiolucent) seen in mammography and the Breast Imaging Reporting and Data System (BI-RADS) lexicon has classified mammographic density into four categories, with the percentage of each tissue density in the general screening population estimated as follows: 10% of women have breasts that are almost entirely fatty, 40% have scattered areas of fibroglandular density, 40% have heterogeneously dense breasts, and 10% have extremely dense breasts.1...
Results
The model accuracy, agreement using cohen’s kappa and derived statistics were calculated using the Statistical functions module from the SciPy 1.10.0 Python Library.
The performance of the model was evaluated on a dichotomous classification, distinguishing BI-RADS a or b (non-dense breasts) from c or d (dense breasts) categories and on the ability to differentiate between the four BI-RADS density categories (a to d), with the results being presented in table 1 (dichotomous classification) and table 2 (four BI-RADS density classification).
[Table 1]
[Table 2]
Based...
Conclusion
The results of this study suggest that while the tool demonstrates a relatively high level of accuracy compared to the original radiologist's density assessment in distinguishing between dense and non-dense breasts, it may have limitations in accurately classifying the specific BI-RADS density categories. This can be seen from the lower mean accuracy of 56.7% and mean agreement of 0.325 when distinguishing between the four BI-RADS categories.
Pronounced variation of accuracy scores was found regarding individual radiologist assessments and the algorithm prediction, highlighting the importance of...
Personal information and conflict of interest
J. Abrantes:
Nothing to disclose
M. J. N. Bento e Silva:
Nothing to disclose
J. P. Meneses:
Nothing to disclose
C. Oliveira:
Nothing to disclose
F. M. G. F. Calisto:
Nothing to disclose
R. W. Filice:
Nothing to disclose
References
ACR BI-RADS atlas: breast imaging reporting and data system ; mammography, ultrasound, magnetic resonance imaging, follow-up and outcome monitoring, data dictionary. (ACR, American College of Radiology, 2013).
Boyd, N. F., Martin, L. J., Yaffe, M. J. & Minkin, S. Mammographic density and breast cancer risk: current understanding and future prospects. Breast Cancer Res. 13, 223 (2011).
Sickles, E. A. The Use of Breast Imaging to Screen Women at High Risk for Cancer. Radiol. Clin. North Am. 48, 859–878 (2010).
Freer, P. E. Mammographic Breast Density:...