Performance of CADx on a large clinical database of FFDM images

Abstract

The purpose of this study is to evaluate the performance of computer-aided diagnosis (CADx) methods for use with images from full-field digital mammography (FFDM) for breast mass lesion classification. A total of 739 FFDM images, including 287 breast mass lesions, were retrospectively collected under an institutional review board approved protocol. All mass lesion margins were delineated by an expert breast radiologist and were used, along with the pathology, as truth in the subsequent evaluation. Our computerized image analysis method for radiologist-indicated lesions consists of several steps: 1) automatic extraction of the lesion from the parenchymal background using computerized segmentation methods; 2) automatic extraction of various lesion features (mathematical descriptors) from image data of the lesions and surrounding tissues; and 3) merging of selected features into an estimate of the probability of malignancy using a Bayesian artificial neural network classifier. The features were selected using a stepwise feature selection procedure. Performance of the CADx system in the task of differentiating between malignant and benign lesions was evaluated using receiver operating characteristic (ROC) analysis. An AUC value of 0.83 was obtained in the task of distinguishing between malignant and benign mass lesions in a leave-one-out by case evaluation with dual-stage segmentation method on the entire FFDM dataset. Results show that the computerized analysis methods for the diagnosis of breast lesions on FFDM are promising, and can potentially be used to aid clinicians in the diagnostic interpretation of FFDM. © 2008 Springer-Verlag Berlin Heidelberg.