Sensitivity and specificity of computer-assisted breast cancer detection in mammography screening

Abstract

Purpose: To evaluate a system of computer-assisted diagnosis (CAD) in mammography.

Material and Methods: A sample of 120 sets of two-view mammograms was examined by an expert screener, a screening radiologist, a clinical radiologist, and a CAD system. The screening and clinical radiologists examined the mammograms twice, first without and then with the help of CAD. The sample consisted of first-round screening films from a two-round population-based screening, and comprised: 32 women in whom breast cancer was detected at the first screening; 10 with cancer detected during the screening interval; 32 with cancer detected at the second screening; and 46 with normal mammograms at both screenings.

Results: The expert screener, the screening radiologist, the clinical radiologist, and the CAD system detected respectively 44, 41, 34 and 37 cancers. Their respective specificities were 80%, 83%, 100% and 22%. With the help of CAD, the screening radiologist detected 1 additional cancer and the clinical radiologist detected 3; their respective specificities were 80% and 100%.

Conclusion: The sensitivity of the CAD system was satisfactory. The two radiologists helped by CAD achieved a modest increase in sensitivity with unaffected specificity. However, the CAD system by itself had a very low specificity and it needs improvement before it can be useful in mammographic screening.