SSE01-05

Predicting Masking Risk in Mammography

Monday, Nov. 26 3:40PM - 3:50PM Room: E451B



Participants
James G. Mainprize, PhD, Toronto, ON (Presenter) Institutional research agreement, General Electric Company
Olivier Alonzo-Proulx, Toronto, ON (Abstract Co-Author) Institutional research agreement, General Electric Company
James Patrie, MS, Charlottesville, VA (Abstract Co-Author) Nothing to Disclose
Jennifer A. Harvey, MD, Charlottesville, VA (Abstract Co-Author) Stockholder, Hologic, Inc; Research Grant, Volpara Health Technologies Limited; Stockholder, Volpara Health Technologies Limited;
Martin J. Yaffe, PhD, Toronto, ON (Abstract Co-Author) Research collaboration, General Electric Company; Shareholder, Volpara Health Technologies Limited; Co-founder, Mammographic Physics Inc; Research Consultant, BHR Pharma LLC

For information about this presentation, contact:

james.mainprize@sri.utoronto.ca

PURPOSE

Masking in mammography is the reduction of lesion conspicuity by surrounding and overlying dense tissue. Masking risk is increased in dense breasts, leading to reduced sensitivity of breast screening. We have developed a masking index that can predict the likelihood of a masked or missed cancer and could be used in a screening program to stratify women at greatest risk of masking to alternative or supplementary imaging modalities to mammography.

METHOD AND MATERIALS

The study population were cancer cases collected (2003-2013) a case-control study used to develop a breast cancer risk model incorporating density measures. Cancers were classified as screen-detected cancers (SDC) found on a screening mammogram and non-screen detected cancers (NSDC) found by clinical symptoms or other imaging. The study had ethics board approval with informed consent All SDC found on baseline images were excluded. Inclusion as NSDC required at least one prior negative screening exam within two years of diagnosis. Images were analyzed with in-house algorithms and by volumetric breast density (VBD) software. The aim in this study was to create an index that differentiated mammograms which allowed for detection (SDC) from those for causing masking or missed lesions (NSDC). To avoid the influence of the lesion itself, only the contra-lateral breast images were used.

RESULTS

The study included 90 NSDC cases and 186 SDC controls. Univariate masking indices based on BMI, age, BI-RADS density, VBD or mean detectability yielded areas under ROC (AUC) of 0.61, 0.65, 0.67, 0.72 and 0.75 (0.06 95% confidence) respectively. For cancers found within one year, the detectability AUC improved to 0.81.

CONCLUSION

Age and BMI are relatively weak predictors of masking risk whereas VBD and detectability measures have better performance. Further, adding textural measures improve predictions slightly, suggesting that the masking effects of anatomic noise and texture are informative. In future, we will validate the model in an independent population and test the result on normal mammograms to predict impact on a stratified screening program.

CLINICAL RELEVANCE/APPLICATION

A reliable masking index to predict when mammography will underperform would be a valuable tool in a stratified screening program which could be used to redirect women with highly masked mammograms to alternative or adjunct screening strategies such as tomosynthesis, MR or ultrasound.