CLAS, a randomized, controlled clinical trial testing niacin/colestipol plus diet therapy in non-smoking men with previous coronary bypass surgery, has demonstrated treatment benefits using both a) a coronary endpoint determination by a panel of human readers, the global coronary change score, and also using b) quantitative coronary angiographic (QCA) measures (percent stenosis, roughness, percent involvement, and minimum diameter). However, in correlating these QCA measures with the global coronary change score using linear regression techniques, only 36% of the variability in the global coronary change score was explained. In this proposal, we apply the technique of adaptive neural networks (ANN) to these QCA measures, as well as measures describing the geometric and hemodynamic relationship between arterial and bypass graft segments, in predicting the panel-based assessment of coronary status. The resulting ANN will be compared to traditional statistical models. In other applications, ANNs have a) demonstrated the potential to recognize complex patterns (such as changes in lesions due to treatment); b) are able to generalize from the available data, enabling appropriate responses to new combinations of input data; and c) are tolerant to missing input data.
Lapuerta, P; Azen, S P; LaBree, L (1995) Use of neural networks in predicting the risk of coronary artery disease. Comput Biomed Res 28:38-52 |