K Hukkinen1, M Pamilo. 1. Department of Radiography, Helsinki University Central Hospital, Hospital District of Helsinki and Uusimaa, Helsinki, Finland. katja.hukkinen@hus.fi
Abstract
PURPOSE: To evaluate whether breast cancers detected at screening are visible in previous mammograms, and to assess the performance of a computer-aided detection (CAD) system in detecting lesions in preoperative and previous mammograms. MATERIAL AND METHODS: Initial screening detected 67 women with 69 surgically verified breast cancers (Group A). An experienced screening radiologist retrospectively analyzed previous mammograms for visible lesions (Group B), noting in particular their size and morphology. Preoperative and previous mammograms were analyzed with CAD; a relatively inexperienced resident also analyzed previous mammograms. The performances of CAD and resident were then compared. RESULTS: Of the 69 lesions identified, 36 were visible in previous mammograms. Of these 36 "missed" lesions, 14 were under 10 mm in diameter and 29 were mass lesions. The sensitivity of CAD was 81% in Group A and 64% in Group B. Small mass lesions were harder for CAD to detect. The specificity of CAD was 3% in Group A and 9% in Group B. Together, CAD and the resident found more "missed" lesions than separately. CONCLUSION: Of the 69 breast cancers, 36 were visible in previous mammograms. CAD's sensitivity in detecting cancer lesions ranged from 64% to 81%, while specificity ranged from 9% to as low as 3%. CAD may be helpful if the radiologist is less subspecialized in mammography.
PURPOSE: To evaluate whether breast cancers detected at screening are visible in previous mammograms, and to assess the performance of a computer-aided detection (CAD) system in detecting lesions in preoperative and previous mammograms. MATERIAL AND METHODS: Initial screening detected 67 women with 69 surgically verified breast cancers (Group A). An experienced screening radiologist retrospectively analyzed previous mammograms for visible lesions (Group B), noting in particular their size and morphology. Preoperative and previous mammograms were analyzed with CAD; a relatively inexperienced resident also analyzed previous mammograms. The performances of CAD and resident were then compared. RESULTS: Of the 69 lesions identified, 36 were visible in previous mammograms. Of these 36 "missed" lesions, 14 were under 10 mm in diameter and 29 were mass lesions. The sensitivity of CAD was 81% in Group A and 64% in Group B. Small mass lesions were harder for CAD to detect. The specificity of CAD was 3% in Group A and 9% in Group B. Together, CAD and the resident found more "missed" lesions than separately. CONCLUSION: Of the 69 breast cancers, 36 were visible in previous mammograms. CAD's sensitivity in detecting cancer lesions ranged from 64% to 81%, while specificity ranged from 9% to as low as 3%. CAD may be helpful if the radiologist is less subspecialized in mammography.
Authors: Elodia B Cole; Zheng Zhang; Helga S Marques; R Edward Hendrick; Martin J Yaffe; Etta D Pisano Journal: AJR Am J Roentgenol Date: 2014-10 Impact factor: 3.959