[Computer-aided diagnosis in the detection of simulated clustered microcalcifications on mammography]

The usefulness of a computer-aided diagnosis (CAD) using a mathematical morphology (Tophat method) for the detection of simulated clustered microcalcifications on mammography (MMG) was evaluated with an acrylic phantom as well as a specially made phantom breast. The sites of microcalcifications indicated by CAD were printed as dots on the films. With the acrylic phantom, CAD detected microcalcifications as small as 0.177 mm, but with the specially made phantom breast the sensitivity decreased from 100% for microcalcifications of 0.250 mm to 25% for those of 0.210 mm. The detectability of microcalcifications was influenced by the contrast with background opacity. A receiver-operating-characteristic (ROC) study performed on 80 images of the phantom breast by 8 radiologists showed no significant difference between the results with and without CAD assistance. However, when the observers were divided into two groups according to the results of interpretation without CAD assistance, the detectability of microcalcifications with CAD significantly improved in the group with the poorer performance. The detection of microcalcifications was influenced by contrast with the surrounding substance in the phantom study. The usefulness of this CAD method presented for the detection of microcalcifications on MMG seems to be limited for inexperienced observers, but it may have potential to be widely applicable with further refinement of the technique.