Campus Access Only

All rights reserved. This publication is intended for use solely by faculty, students, and staff of University of the Pacific. No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, now known or later developed, including but not limited to photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author or the publisher.

Date of Award

2005

Document Type

Thesis - Pacific Access Restricted

Degree Name

Master of Science (M.S.)

Department

Pharmaceutical and Chemical Sciences

First Advisor

Timothy J. Smith

First Committee Member

Sian Carr-Lopez

Second Committee Member

Roshanak Rahimian

Abstract

Artificial Neural Networks are biologically inspired computational methodologies that can perfom multifactorial analyses. In recent years, they have been evaluated for medical decision support, with varying degrees of success. The preliminary part of my thesis deals with evaluating whether an Artificial Neural Network can be trained to approximate a cardiovascular risk stratification algorithm by Rifai et al. My subsequent work involves training the network on a population-based cross-sectional dataset with the objective of categorizing Ankle-Brachial Index and Maximal Oxygen Consumption. These are indicators of the severity of lower extremity atherosclerosis and the level of cardiovascular fitness respectively.

NeuralSIM®, a commercially available Artificial Neural Network, was trained using C-reactive protein and Total Cholesterol/HDL Cholesterol ratio as input parameters, and the relative risk stratum for future myocardial infarctions or stroke as output. For the Ankle-Brachial Index and the cardiovascular fitness networks, data was obtained from the National Health and Nutrition Examination Survey. The network for cardiovascular fitness was compared with an algorithm published by Jackson et al.

The network was able to approximate the cardiovascular risk stratification algorithm by Rifai et al closely with correlation coefficients of0.95 in men and 0.93 in women respectively. The network to screen for low cardiovascular fitness had a sensitivity of 83% and a specificity of 78%, with an overall accuracy of 81%. The Ankle-Brachial Index network demonstrated a high level of specificity (86.3%) for estimating abnonnal values but a very low sensitivity (30%).

Artificial neural networks showed encouraging results for potential use as decision-support tools. One significant limitation is that the importance of individual parameters or the exact function cannot be ascertained easily. There is a need to address this issue in future software development.

Pages

50

To access this thesis/dissertation you must have a valid pacific.edu email address and log-in to Scholarly Commons.

Find in PacificSearch

Share

COinS

If you are the author and would like to grant permission to make your work openly accessible, please email