Analysis of false-negative readings of automated breast ultrasound studies

Background: To assess the reasons for false-negative readings of automated breast ultrasound (ABUS) studies. Methods: Between 2012 and 2015, 1,890 ABUS studies were performed at our tertiary medical center. Those for which false-negative results were documented in the initial ABUS report against the corresponding hand-held ultrasound (HHUS) scan were reviewed by three specialized breast radiologists. Key images of specific lesions were marked on the ABUS and HHUS scans and compared for quality (equal, better with HHUS, better with ABUS). Readers were also asked to identify the reasons for the differences in image quality between the scans: poor visibility, lesion location, or fibroglandular tissue shadowing. Results: Twenty-two ABUS studies met the study criteria. Two of the three readers found that most lesions were better demonstrated with HHUS. Overall agreement among the readers was moderate (kappa 0.36, SD 0.15, p = 0.002). Highest agreement was found for better image quality for HHUS than ABUS (kappa 0.4, SD 1.3, p = 0.0007). Of the four biopsy-proven carcinomas, three were found by all three readers to be better depicted with HHUS; two were located peripherally and were not seen by ABUS. For all readers, the most common reason for false-negative readings was poor visibility, followed by peripheral lesion location and shadowing obscuring the lesion. Conclusions: Several factors may make reading ABUS images difficult. Resolution can be diminished by imperfect transducer-breast contact, and fibrotic breasts can cause artifacts such as marked shadowing. Peripheral lesions may be missed because of blind spots. Reader training and experience may play an important role in managing these issues.