Related Experiment Video
Updated: Jun 27, 2026
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Does computer-aided detection have a role in the arbitration of discordant double-reading opinions in a
1Nottingham Breast Institute, Nottingham City Hospital, Nottingham, UK. jonathan.james@nuh.nhs.uk
Aims:
To investigate whether a computer-aided detection (CAD) system could act as an arbitrator of discordant double-reading opinions, replacing the need for an independent third film reader.
Methods:
The mammograms of the 240 women that underwent arbitration by an independent third reader were identified from the 16,629 women attending our screening centre between July 2003 and April 2004. Mammograms of the arbitration cases were digitized and analysed by a CAD system. To assess the ability of CAD to act as the arbitrator, the site of the CAD prompts was retrospectively compared to the site of any abnormality noted by the original film readers. If a CAD prompt was placed on a region marked by one of the film readers then the decision of CAD as the arbitrator was that the women should be recalled for further assessment. If no mark was placed then the region was considered low risk and the decision was not to recall. The decision of CAD as the arbitrator was retrospectively compared with the original recall decision of the independent third reader.
Results:
There were 21 cancer cases in the group of women undergoing arbitration, diagnosed both at the original screening episode and subsequently. The independent third reader recalled 15/18 (83%) of the cancers that corresponded with the arbitrated lesion. CAD as the arbitrator would have recalled 16/18 (89%) of the cancers that corresponded to the arbitrated lesion. CAD acting as the arbitrator would have resulted in a significant increase in normal women being recalled to assessment in the arbitration group (P<0.001). The extra 50 recalls would have potentially increased the overall recall rate to assessment from 3.1 to 3.4%; a relative increase of 10%.
Conclusions:
The main effect of CAD acting as an arbitrator of discordant double-reading opinions is to increase the recall rate, significantly above what is found when arbitration is performed by an independent third reader. Using CAD as an arbitrator may be an option to deal with discordant double-reading opinions when no other method of consensus or arbitration is available.
