Can digital breast tomosynthesis accurately predict whether circumscribed masses are benign or malignant in a screening population?

Abstract

Aim: To evaluate whether digital breast tomosynthesis (DBT) can predict if circumscribed masses are benign or malignant by assessing margin sharpness.

Materials and methods: Circumscribed masses were evaluated on co-registered two-dimensional digital mammography (2DDM) and DBT. Lesions were categorised as follows: category 1=visible sharp border 0-25% of the total margin; category 2 = 26-50% category 3= 51-75%, and category 4=76-100%. Changes in category between 2DDM and DBT were analysed; if the category was lower on DBT the change was negative, if higher the change was positive.

Results: Of 759 lesions, 121 masses classified as circumscribed on DBT were included; 25 were malignant and 96 benign. Of the benign lesions, 8/96 were within category 3 or 4 on 2DDM compared with 48/96 benign lesions within category 3 or 4 on DBT (Fisher's exact test p<0.000527). Forty-eight of 51 (94.1%) lesions categorised as 3 or 4 on DBT were benign and 65/67 (97.01%) of the positive category change group were benign. Lesions in category 1 on DBT had 45.4% chance of being malignant (20/44) compared with 22.72% (20/88) on 2DDM (chi-squared test p<0.001). Sixty-five of 67 (97.01%) lesions in the positive category change group were benign and 23/54 (42.6%) lesions with either no or negative category change were malignant.

Conclusion: The present study demonstrates 97% accuracy in predicting circumscribed lesions as benign when using positive category change and 94% accuracy when >50% of the margin is sharply defined on DBT.