Purpose: Contrast-enhanced mammography (CEM) is a functional mammographic technique that improves the diagnostic accuracy of digital mammography (DM) with comparable sensitivity and increased specificity relative to breast MRI. DM is suboptimal for differentiating benign and metastatic lymph nodes (LN) due to its coverage limitations and low accuracy. We evaluate the potential added value of subtracted contrast images (SCI) for differentiating malignant versus benign LN visible on staging CEM.
Materials and Methods: A retrospective multi-reader evaluation was performed on patients with untreated invasive breast cancer (BC) or DCIS, or highly suspicious imaging findings (BI-RADS 4C or 5), who enrolled in a prospective IRB-approved protocol. The patients underwent bilateral staging CEM using a DM/DBT mammography system with CEM capabilities (Senographe Pristina, GE Healthcare, Buc, FR), followed by bilateral breast and ipsilateral axillary LN ultrasound (US). Patients with axillary LN visible on CEM and the ground truth of the nodal status established by US-guided biopsy, sentinel lymph node biopsy, or positive PET findings, were included in this study. Seven fellowship-trained breast radiologists (6-23 years of experience) with no CEM experience, but proficient in other imaging modalities, evaluated cropped low energy (LE, DM equivalent) and SCI images of LN in random order without reference to mammography or US imaging, and were blinded to the ground truth. Three separate datasets of LN images were read sequentially: 1) LE; 2) SCI; 3) combined LE/SCI. Image interpretation was dichotomized as positive (BI-RADS 0,4,5) versus negative (BI-RADS 1 and 2) with no option for BI-RADS 3. Descriptive statistics was used for data analysis.
Results: 27 axillary and intramammary LN on 22 patients were selected (19 benign, 8 metastatic). Reader accuracy ranged from 0.40- 0.70 for LE; 0.48- 0.81 for both SCI and LE/SCI, with mean accuracy of 0.54 for LE, 0.63 for SCI, 0.63 for LE/SCI (Fig. 1, Table 1). The addition of SCI improved overall accuracy in the majority of the readers (range 0- 18%, mean 9%). In one instance, there was 100% agreement among radiologists on discrepant LE and SCI interpretation (Fig. 2A, 2B).
Conclusion: Addition of SCI mildly improves the diagnostic accuracy of DM for assessment of LN status, but the overall accuracy remains low with a wide range of accuracies, similar to DM. This is likely modality and/or technique rather than reader dependent and renders CEM suboptimal for this purpose.
Clinical Relevance Statement: CEM is suboptimal for LN interpretation, and ultrasound remains the modality of choice.
