COMPUTER AIDED DIAGNOSIS OF BREAST CANCER INVASION

Duke University, Durham NC

Investigators

Linked publications & trials

Abstract

DESCRIPTION: The purpose of this study is to develop a computer-aided diagnosi (CADx) system to predict breast lesion malignancy and invasion based on medica findings. Artificial neural network (ANN) techniques will be used to predict whether mammographically suspect lesions are benign, in situ cancer, or invasive cancer. The ANN inputs will be derived from existing, available information such as patient history and radiologist[unreadable]s descriptions of lesion morphology following the ACR Breast Imaging Reporting and Data System (BI-RADS). ANNs are well suited for this diagnostic task because, like humans, ANNs can be taught to perform diagnostic tasks accurately and robustly when given appropriate training examples. The specific aims of the proposed study are to: (1) Develop ANNs that use mammography and history findings to predict malignancy and invasion of breast lesions among a prospectively collected patient database; (2) Refine the accuracy of the CADx system by optimizing the number of input findings and investigating more complex network architectures, and study is cost-effectiveness. (3) Evaluate the CADx system clinically, by developing a graphical user interface and using it to retrospectively evaluate the system[unreadable]s performance. In preliminary studies, an ANN accurately predicted invasion among 96 biopsy-proven breast cancers, using BI-RADS findings and patient age as input findings. The immediate benefit of this proposal is a noninvasive computer-aided diagnosis system which provides information previously available only through biopsy. This system can assist mammographers and surgeons in surgical planning for patients with breast lesions, and may reduce the cost and morbidity of unnecessary surgical biopsies.