Application of a new breast phantom in the evaluation of computer-aided diagnosis schemes

Abstract

The availability of computational tools - from simplest to most complex - to aid in diagnostic radiology is already a reality, becoming in many cases an intrinsic resource of the process. In the field of mammography, these tools have been gathered into schemes called computer-aided detection / diagnosis (CADe / CADx), which have been proposed in order to reduce the difficulties of evaluating mammographic images. However, these schemes need to be adequately evaluated with a very large collection of images in order assure a statistically significant survey in determining their sensitivity and specificity, which is one of the key factors in ensuring the physician's trust in their results. Thus, the proposal of this project consists in the establishment of a process to evaluate a CADx scheme previously developed from a large set of images to be obtained with a new breast phantom capable of generating several patterns of images by the random variation of size, shape, contrast and distribution of simulated lesions. Once established the required standardization of the evaluation criteria from this new database, the feedback of the CADx scheme under evaluation from the tests results will allow setting up selected processing techniques in order to assure the best possible efficacy with images obtained from different mammography equipment. Hence, the final product will be an application in JAVA language to be used as an aid tool for applications in the diagnostic routine considering any set of digital mammography images in DICOM standard - in addition to the availability of a new breast phantom image database for free access by Internet. (AU)