A Pilot Study to Explore Linkages Among Isomers of Organochlorines, Promutagenic DNA Lesions and Breast Cancer Using Sensitive Techniques
Abstract
The purpose of this grant was to construct an artificial neural network (ANN) to assist radiologists in differentiating benign from malignant solid lesions in ultrasound (U.S.) breast imaging. A data set of patient cases was collected, consisting of 192 biopsy-proven breast lesions for which radiologists provided descriptive terms to characterize the U.S. appearance of the lesions. An ANN model was developed to predict probably benign lesions based upon those descriptors and the patient age. The model was potentially able to maintain 100% sensitivity of cancer detection, while improving the radiologists' specificity from 0% to 35% (42 out of 121 benign biopsies obviated). This corresponded to improving the PPV of the radiologists from 37% to 47%. Moreover, we also identified that the mass margin and patient age were the two most important input features for this model, and that highly simplified models based on those two features alone could still perform as well as the more complicated models using all available information. Predictive models such as these can provide physicians and patients with accurate information for managing suspicious breast lesions without the invasiveness of biopsy procedures.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2000
- Accession Number
- ADA385883
Entities
People
- Joseph Y. Lo
Organizations
- Duke University Hospital