Dr. Lingyun Bao from the First People's Hospital of Hangzhou in Zhejiang and colleagues analyzed 1,000 ABUS cases that had been performed in 2012. Of these cases, 206 were malignant, 486 were benign, and 308 were normal; all the malignant and benign cases were confirmed by biopsy or surgery, while normal cases were confirmed by two-year follow-up.
Nine radiologists read two batches of cases that were randomly assigned. Initially, the first batch was read with the help of the CAD system and the second batch without. After a month, the radiologists reread the first batch of cases without CAD and the second batch with the software. Bao's group compared the area under the curve (AUC) of the CAD-aided readings to those read without CAD, and they recorded the reading time per case for each radiologist.
The AUC was 0.784 for all readers using CAD with ABUS and 0.747 for unassisted reads, a statistically significant difference, the group reported. The average reading time was 10% faster when CAD was used, leading the researchers to conclude that the software improves radiologist performance in both accuracy and reading time with ABUS.
