The order of mammography reading with respect to breast density may facilitate improved workflow efficiency and reduced reading time for radiologists, according to a new study.
For the retrospective study, recently published in Radiology, researchers at the possible impact of the order of mammography worklist review by comparing an order via increasing volumetric breast density (VBD), a random order and ordering through self-supervised learning (SSL), which entailed automated grouping of similar looking mammography images. Thirteen radiologists reviewed mammograms from a cohort of 150 women (median age of 55), evenly divided between those with and without breast cancer, according to the study.
In comparison to a random reading order, reading in order of increasing VBD demonstrated no significant differences in sensitivity (81 percent for both) nor specificity (86 percent vs. 89 percent), and a slight increase in area under the curve (AUC) (92 percent vs. 93 percent). However, the researchers noted that the VBD order led to reduced mean reading time (24.3 seconds vs. 27.9 seconds) and reduced fixation time on malignant regions (3.7 seconds vs. 4.6 seconds) in contrast to random ordering of mammography review.
“By integrating quantitative density metrics into the screening workflow and implementing automated sorting algorithms in the screening worklist, organizing batch readings of screening examinations from low to high VBD would be feasible to implement and may serve as a valuable strategy to optimize breast cancer screening,” wrote lead study author Jessie J.J. Gommers, MSc, who is affiliated with the Department of Medical Imaging at the Radboud University Medical Center in Nijmegen, the Netherlands, and colleagues.
The researchers also noted no significant differences between the SSL order versus random order of mammography review for sensitivity (80 percent vs. 81 percent), specificity (84 percent vs. 86 percent) and AUC (92 percent for both).
There was also no difference between the three ordering approaches with respect to time to first fixation (TTFF) for lesions, according to the study authors. In light of these findings, the study authors said the combination of visual adaptation and improved lesion characterization could be a key driver in the enhanced efficiency with the VBD ordered mammography.
“Visual adaptation may play a role in differentiating between normal versus unexpected characteristics in images. However, lesion characterization involves more complex cognitive processes, such as pattern recognition, knowledge of lesion characteristics, and clinical experience, as well as metacognition or confidence in judgments, which may be influenced by adaptation,” explained Gommers and colleagues.
Three Key Takeaways
1. Reading mammograms in order of increasing breast density may improve efficiency. Organizing mammography readings by increasing volumetric breast density (VBD) showed reduced mean reading time and decreased fixation time on malignant regions, suggesting it may help optimize the screening workflow without compromising diagnostic accuracy.
2. No significant differences in diagnostic performance. In comparison to random ordering, ordering by VBD did not lead to significant changes in sensitivity or specificity, but there was a slight improvement in the area under the curve (AUC).
3. Potential for broader application in breast imaging. Although no differences were found between the VBD ordering, self-supervised learning (SSL) ordering and random ordering with respect to time to first fixation (TTFF) or lesion detection, integrating VBD into workflows could be further researched, especially for digital breast tomosynthesis (DBT), which involves longer reading times.
In an accompanying editorial, Lars J. Grimm, M.D., MHS, praised the researchers for demonstrating the enhanced efficiency of mammography review with the increasing VBD order. Dr. Grimm noted that he would like to see further research of this ordering approach and its possible impact on digital breast tomosynthesis (DBT).
“ … There is the potential for much greater efficiency gains with DBT since the average reading time for DBT can be twice as long as that for 2D mammography. Logistically, interpretation time is one of the major burdens that high-volume breast imaging practices are dealing with as they transition to DBT. Thus, any improvements would be well received,” emphasized Dr. Grimm, an associate professor in the Department of Radiology at Duke University.
(Editor’s note: For related content, see “FDA Clears New Features in AI-Powered Mammography Software Suite,” “Radiology Experience, Breast Density and Screening Mammography: What New Research Reveals” and “Study: Contrast-Enhanced Mammography Offers Significantly Higher Sensitivity for Breast Cancer in Dense Breasts.”)
In regard to study limitations, the authors acknowledged the cohort was derived from a cancer-enriched case set without access to prior mammograms of the patients. They also conceded an emphasis on visual location of lesions by radiologists for some cases in which biopsy information was unavailable.
