RSNA 2008 

Abstract Archives of the RSNA, 2008


SSM02-03

CAD for Mammography Made More Effective for Small Lesions in Dense Breasts by Filtering the Detection Marks by Their Level-of-Suspicion

Scientific Papers

Presented on December 3, 2008
Presented as part of SSM02: Breast Imaging (Computer-aided Detection)

Participants

Issac Leichter, Presenter: Employee, Siemens AG
Richard Lederman MD, Abstract Co-Author: Research Consultant, Siemens AG
Eli Ratner, Abstract Co-Author: Employee, Siemens AG, Jerusalem, Israel
Nicolas Merlet, Abstract Co-Author: Employee, Siemens AG
Glenn Fung PhD, Abstract Co-Author: Employee, Siemens AG
Balaji Krishnapuram PhD, Abstract Co-Author: Employee, Siemens AG

PURPOSE

To investigate the impact of user-controlled filtering of CAD marks, on the detection of small invasive masses in dense breast, which are most difficult to identify but greatly affect the prognosis of breast cancer.

METHOD AND MATERIALS

149 pathology-proven FFDM cases with malignant masses and 528 normal cases were culled retrospectively, in a consecutive manner, from 3 screening clinics. All cases were run on a prototype CAD device (Siemens) that used quantitative features characterizing each potential finding to calculate a certainty score, reflecting the level-of–suspicion of the candidate. The device displayed only candidates with a certainty score above a selected threshold, and the user could toggle between thresholds and analyze in real time, which CAD marks are filtered at each level. A non-blinded radiologist recorded the location of the biopsied finding, breast density category and lesion size. Lesions smaller than 15mm were considered small. Detection was assessed by comparing each CAD mark to the biopsied finding. The detection sensitivity and the false mark (FM) rate at 3 filtering levels were analyzed by lesion size and breast density.

RESULTS

Lesion size affected the detection sensitivity for masses much more than breast density. For all filtering levels, the FM rate was substantially lower in normal cases than in malignant cases. With the highest level of filtering, the mass FM rate in the malignant cases decreased by 37.7%, and in the normal cases, by 49.1%. This significant reduction was associated with a decrease of only 4.4% in the detection sensitivity for small masses in dense breasts (from 69.6% to 65.2%), and with a similar decrease of 4.7% in sensitivity (from 84.8% to 80.1%) for all masses.

CONCLUSION

Surprisingly, increasing the level of filtering of the CAD marks does not have a selective adverse impact on the detection sensitivity of difficult cases with small masses in dense breast. A CAD device with user-controlled filtering of CAD marks, is more effective since the substantial reduction in false marks offsets the slight reduction in the detection sensitivity, even for small masses in dense breasts.   

CLINICAL RELEVANCE/APPLICATION

The filtering of CAD prompts reduces false marks without a selective decrease in the detection sensitivity of small malignant masses in dense breasts, where CAD should have its greatest impact.  

Cite This Abstract

Leichter, I, Lederman, R, Ratner, E, Merlet, N, Fung, G, Krishnapuram, B, CAD for Mammography Made More Effective for Small Lesions in Dense Breasts by Filtering the Detection Marks by Their Level-of-Suspicion.  Radiological Society of North America 2008 Scientific Assembly and Annual Meeting, February 18 - February 20, 2008 ,Chicago IL. http://archive.rsna.org/2008/6020522.html