Computer-Aided Diagnosis and Automated Screening of Digital Mammogram

摘要

We have developed image analysis software which is capable of detectingmammographic abnormalities which would be used in a second reader scenario to prompt a radiologist to more carefully analyze suspicious regions in the mammogram. For an average of about 2 prompts per image, our algorithm detected 70% of the lesions in our database. In a second reader scenario, it is only necessary to detect a lesion in at least one of the two views. Using this criteria, our algorithm detected 90% of the masses, and perhaps more importantly 97% of the malignant masses were found. The automated prompting and the additional information provided by computerized image analysis should result in greater repeatability and uniformity in the standard of care. As several recent studies have indicated, it should also result in some increase in sensitivity for a given level of specificity. The extra cancers detected would then be treated earlier and less expensively.