Performance of computer-aided diagnosis in the interpretation of lesions on breast sonography

摘要: Abstract Rationale and objectives To investigate the potential usefulness of computer-aided diagnosis as a tool for radiologists in characterization classification mass lesions on ultrasound. Materials methods Previously, computerized method automatic breast ultrasound was developed. The includes segmentation lesion from image background extraction four features related to shape, margin, texture, posterior acoustic behavior. In this study, effectiveness computer output an aid their ability distinguish between malignant benign lesions, patient management decisions terms biopsy recommendation are evaluated. Six expert mammographers six private practice at institution accredited by American Ultrasound Institute Medicine participated study. Each observer first interpreted 25 training cases with feedback results, then 110 additional without feedback. Simulating actual clinical setting, were unknown both observers computer. During interpretation, gave confidence that also (biopsy or follow-up). displayed, again recommendation. Statistical analyses included receiver operator characteristic analysis Student t -test. Results For community radiologists, A z (area under curve) increased 0.83 0.87 ( P = .02) 0.80 0.84 .04), respectively, when used interpretation images. Also, values similar value alone (A 0.83). Conclusion Computer images has been shown improve diagnostic accuracy task distinguishing recommending biopsy.