Interreader variability and predictive value of US descriptions of solid breast masses: pilot study.

Abstract

RATIONALE AND OBJECTIVES: This study evaluated the specificity of ultrasound (US) characteristics of solid breast lesions and the interreader variability in their interpretation. MATERIALS AND METHODS: In 61 patients, 70 sonographically visible solid masses, scheduled for biopsy because of findings from conventional imaging, were prospectively and sequentially accrued for evaluation. Three readers interpreted the sonograms and described the solid masses in terms of established US characteristics. The specificity and positive predictive value (PPV) for each characteristic were calculated by comparing US findings with biopsy findings, and interreader variability was evaluated. Five assessment categories were developed to guide recommendations for patient care. The relative performance of each reader was assessed by measuring the PPV for each assessment category and by measuring the area under the receiver operating characteristic curve. RESULTS: The specificity and PPV were calculated for all characteristics and for each reader. The average specificities of the three readers for the most frequently used six characteristics were as follows: spiculation, 97%+/-5 (standard deviation); taller than wide, 91%+/-4; central shadowing, 77%+/-1; markedly hypoechoic, 86%+/-5; duct extension, 95%+/-5; and microlobulation, 84%+/-3 (overall average specificity, 88.5%). The average PPVs for categories II-V were 5%, 10%, 63%, and 94%, respectively. The readers' interpretations were similar and correlated well. CONCLUSION: The proposed US recommendation system is an accurate predictor of histologic findings. A sonographic classification lexicon should prove valuable.