Authors' objectives:

This review aims to assess the available evidence on the computer aided detection of breast cancer in mammography images.

Authors' recommendations: With regard to computer aided detection (CAD) of breast cancer, the ICSI Technology Assessment Committee finds: CAD causes no direct harm to patients. Misdiagnoses (false-positive or false-negative results) can lead to unnecessary follow-up testing, unnecessary patient anxiety, or delays in obtaining treatment. Lower numbers of false positive marks per image are typically observed for microcalcifications (0.20-0.30 marks/image) compared with masses (0.75-1.0 marks/image). In two studies involving screening populations, between 10% and 16% of malignancies were detected by CAD alone and not detected by a radiologist. In one of these studies involving over 12,800 women, the detection rate for cancer increased 19% (from 3.2 to 3.8 cases per 1,000) with the addition of CAD. The evidence supports the use of CAD or a second reading by an independent radiologist to assure quality results. For microcalcifications, the reported sensitivity of CAD is 98% to 100%. For masses, architectural distortions, and densities, the reported sensitivity is 67% to 90%. (Conclusion Grade II) The radiologist remains the reader and interpreter of the mammogram. CAD assists the radiologist by identifying areas warranting further review. In one study with 208 lesions, masses of 10 to 30 mm in diameter were more likely to be detected by a CAD system compared with either smaller or larger lesions. The recall rate increased only slightly with the addition of CAD. Reported values ranged from 6.5% to 14.4% without CAD and 7.6% to 15.8% with CAD. The positive predictive value for predicting malignant histopathology on the biopsy was unchanged. Reported values ranged from 28% to 38%. Data from one study suggest that CAD applied to digital mammography approximates the results achieved when CAD is applied to film-screen images.

Authors' methods: Review

Project Status: Completed

URL for project: http://www.icsi.org/knowledge/detail.asp?catID=107&itemID=1594

Year Published: 2004

English language abstract: An English language summary is available

Publication Type: Not Assigned

Country: United States

Organisation Name: Institute for Clinical Systems Improvement

Contact Address: 8009 34th Avenue South, Suite 1200, Bloomington, MN, USA. Tel: +1 952 814 7060; Fax: +1 952 858 9675

Contact Name: icsi.info@icsi.org

Contact Email: icsi.info@icsi.org

Copyright: Institute for Clinical Systems Improvement (ICSI)