Alan P. Zelicoff, MD
This content is PDF only. Please click on the PDF icon to access.
To the Editors: Baxt's (1) successful use of neural networks in accurately diagnosing myocardial infarction in an emergency room setting illustrates the vast potential utility of this technique for medical decision making. Unfortunately, despite an apparently straightforward mathematical foundation, much of the design and training of neural networks remains an intuitive art. Network theory provides very little guidance for optimizing network structure, including optimizing the depth of hidden layers and the number of nodes in those layers, which are important determinants of computing resources necessary for training the network (2). Such considerations are paramount in large input nets and in
Zelicoff AP. Myocardial Infarction Prediction by Artificial Neural Networks. Ann Intern Med. ;116:701–702. doi: 10.7326/0003-4819-116-8-701
Download citation file:
Published: Ann Intern Med. 1992;116(8):701-702.
Copyright © 2019 American College of Physicians. All Rights Reserved.
Print ISSN: 0003-4819 | Online ISSN: 1539-3704
Conditions of Use