CHICAGO—Errors in interpretation of ultrasound screening of breast cancer were similar in prevalence (21 percent of misses) to errors in mammographic and MRI interpretation, based on a retrospective review of the ACRIN 6666 trial, presented Nov. 27 at the 97th Scientific Assembly and Annual Meeting of the Radiological Society of North America (RSNA).
The ACRIN [American College of Radiology Imaging Network] 6666 trial enrolled 2,809 women from 21 sites (in the U.S., Canada and Argentina) between April 2004 and February 2006, explained the trial’s principal investigator Wendie A. Berg MD, PhD, visiting professor at the University of Pittsburgh School of Medicine.
In this analysis, Berg et al sought to determine sources of false negative mammography, ultrasound and MRI in the ACRIN 6666 protocol (100 invasive, median size 10 mm). In particular, they sought to assess whether primary factors were lack of depiction or errors in interpretation.
Three experienced breast imaging radiologists independently reviewed breast imaging performed the year prior to and year of diagnosis for each of 130 malignant lesions (110 women) in the 21-site ACRIN 6666 protocol of ultrasound screening. After initial review blinded to other imaging and pathology, observers reviewed pathologic findings and recorded imaging findings evident only in retrospect.
Factors affecting lesion detection included:
- Technical reasons (outside field of view, poor soft tissue contrast, motion or other);
- Detection (“busy breasts,” poor lesion to background contrast, seen on only one view, difficult location and large breast); and
- Interpretation issues (looks benign, multiple similar findings, at scar and very subtle).
The researchers compared the results to initial site interpretations.
For the 52 percent malignancies missed by mammography, Berg reported that 16 were evident on prospective review (five masses, seven calcifications, two calcified masses, one suspicious node and one asymmetry/distortion), three seen only in retrospect (one calcification and two asymmetries) and 48 not depicted. Five of 19 cancers also were seen on prior-year mammogram. Initial mammographic (overlapping) errors were: seven technical, 13 detection and seven interpretation issues.
For the 55 percent malignancies missed by initial ultrasound, 14 were evident prospectively, one only retrospectively and 56 not depicted. Two of 15 cancers were depicted on prior-year ultrasound. Initial ultrasound (overlapping) errors were: six technical, eight detection and nine interpretation factors. Targeted ultrasound depicted another 14 percent malignancies (two not within initial field of view).
Of 19 cancers imaged by MRI, 26 percent were missed on initial site interpretation. Of those five, one mass was seen on prospective review (interpretive error), none in retrospect and four were not depicted.
Based on these results, 28 percent of mammographically missed cancers were interpretive errors as were 21 percent of sonographically missed cancers and 20 percent of MRI misses. Berg reported that another 14 percent missed cancers were seen on targeted ultrasound.
Berg noted that the comparable rate of missed cancers with mammography was “surprising” adding that the “similar rates of errors across all modalities…is important to consider with the improving the performance of ultrasound going forward.”
“Automated scanning will not eliminate errors in interpretation. It is possible, but not proven that computer-assisted detection and diagnosis have reduced errors in mammography and MRI—albeit with some major asterisks.”
She also noted that the results from the ACRIN 6666 trial might be “as good as it is going to get [with the current state of technology] because of the experience level of the readers.” However, she concluded that more study is warranted in this area.
The ACRIN 6666 trial was funded by the Avon Foundation and the National Cancer Institute.