While screening mammography has been shown to be an effective method for the early detection of breast cancer, currently, 5-30 percent of women with breast cancer have a marnmogram that is interpreted as normal. It has been reported that interpretation errors (when the radiologist sees the cancer but reports it as benign) are the cause of 54 percent of missed cancers. In addition, only 10-40 percent of women who have a biopsy actually have breast cancer; with biopsies being expensive, invasive and traumatic to the patient. In addition, there is large inter-observer variability in the interpretation of mammographic lesions. The long-term goal of this research is to develop and evaluate computer-aided diagnosis and prognosis methods for multi-modality imaging of the breast. The main hypotheses to be tested are that, combined information from the computerized analysis of mammography, breast ultrasound, and MR images, along with clinical data, should yield improved methods for (a) distinguishing between malignant and benign lesions, i.e., diagnosis and (b) predicting prognosis. The objectives are to create databases containing mammogram, ultrasound, and MR images along with clinical information, malignant/benign status, and patient outcomes; to develop computerized methods for characterizing the essential morphological, textural, sonographic, and vascular features of the lesions; and to evaluate the accuracy of these methods in distinguishing between malignant and benign lesions and in predicting patient prognosis. It is expected that the results from this research will aid radiologists/oncologists in determining the likelihood of malignancy and in predicting patient prognosis. The proposed work is novel in that such a comprehensive system for computer-aided diagnosis has not yet been attempted. We believe that with the combined information from multimodality imaging and clinical information, overall patient outcome will improve.
Showing the most recent 10 out of 25 publications
