Automated breast ultrasound can significantly augment breast cancer screening attempts completed with mammography. But, even with a second sweep, it’s possible for a radiologist to overlook a finding or misdiagnose it. In those instances, computer-aided detection can be useful.
In a recent study published in Radiology, Hongping Song, MD, PhD, at Xijing Hospital of the Fourth Military Medical Univeristy, and colleagues determined that adding computer-aided detection into reading an automated breast ultrasound image can improve performance for both novice and experienced radiologists. However, the greatest benefit is seen among readers who have fewer years of experience.
“Our study showed computer-aided detection at both the second-reading mode and concurrent-reading mode may improve novice readers’ performance of breast cancer detection at automated breast ultrasound, especially in women who are asymptomatic,” Song wrote in the study.
To determine if and how computer-aided detection impacted a reader’s capability, Song’s team conducted a retrospective study that included 1,485 images collected from 1,452 women. Among the images, 282 had malignant lesions, 695 had benign lesions, and 508 were lesion-free images. Of all the women, with an average age of nearly 44, 529 were healthy with no evidence of breast cancer.
A total of 529 women (36.4%) were asymptomatic—19 had malignant lesions, 85 had benign lesions, and 425 had no lesions. The remaining 923 women (63.6%) were symptomatic with palpable masses, pain, or nipple discharge. All total, 251 women had 263 malignant lesions, 589 women had 610 benign lesions, and 83 were healthy.
Song’s team assigned three novice readers, who had one-to-three years’ experience, and three seasons readers, who had five-to-10 years’ experience, to read images. They completed the work in two sessions scheduled four weeks apart. The team compared the readers’ performance after using computer-aided detection as a second-reading supplemental tool, as well as a concurrent-reading tool.
Related article: Automated Breast Ultrasound Use on the Rise
Overall, the team reported, computer-aided detection was most helpful for inexperienced readers looking at images from asymptomatic women. Improvement was seen both when computer-aided detection was used as part of a second reading and when it was used simultaneously. In particular, both novice and experience readers missed fewer malignancies when using computer-aided detection. Their sensitivity improved from 90% to 94% with second readings and from 90% to 93% with concurrent ones.
In addition, novice readers saw substantial improvement in their sensitivity with the use of BI-RADs category 3 threshold, increasing from 67% without computer-aided detection to 88% without the second and concurrent readings.
Alongside greater sensitivity, adding computer-aided detection into automated breast ultrasound also decreased the time needed to analyze images, according to Song’s team. Based on their analysis, using computer-aided detection with concurrent readings shaved 16 seconds — approximately 32%— off the time spent reviewing the images. The time savings grew to 27 seconds—approximately 44%—with the second readings.
Although computer-aided detection impacted both novice and experienced readers, it did so differently. For the inexperienced readers, computer-aided detection helped them identify more invasive carcinomas than ductal carcinoma in situ. Among the experienced readers, the beneficial impact came with being able to reduce the number of lesions that needed to be sent for biopsy, instead referring more for watching and later follow-up. This impact is particularly important, Song says, due to the low specificity typically associated with supplemental ultrasound screening.
Despite these results that point to a beneficial impact on the reading and diagnostic performance of both novice and experienced radiologists, there is still a need for further research in this area, according to Song’s team. Multi-center prospective studies could be effective in investigating how useful computer-aided detection could be in augmenting breast cancer screening with automated breast ultrasound.