Abstract
ANNs are nonlinear regression computational devices that have been used for over 45 years in classification and survival prediction in several biomedical systems, including colon cancer. Described in this article is the theory behind the three-layer free forward artificial neural networks with backpropagation error, which is widely used in biomedical fields, and a methodological approach to its application for cancer research, as exemplified by colon cancer. Review of the literature shows that applications of these networks have improved the accuracy of colon cancer classification and survival prediction when compared to other statistical or clinicopathological methods. Accuracy, however, must be exercised when designing, using and publishing biomedical results employing machine-learning devices such as ANNs in worldwide literature in order to enhance confidence in the quality and reliability of reported data.
Highlights
Introduction and Development of Artificial NeuralNetworksArtificial neural networks (ANNs) are regression devices containing layers of computing nodes with remarkable information processing characteristics
Advantages include: (a) requirement for less formal statistical training to develop, (b) having a better discriminating power than other regression models, (c) can be developed using multiple different training algorithms, (d) their parallel nature enable them to accept a certain amount of inaccurate data without a serious effect on predictive accuracy, (e) having the ability to accurately detect complex nonlinear relationships between independent and dependent variables, and all possible interactions between variables, as they make no assumptions about those variables, (f) reduce the number of false positives without significantly increasing the number of false negatives, and (g) they may allow for individual case prediction
Disadvantages include: (a) considered as "black box" methods, one cannot exactly understand what interactions are being modeled in their hidden layers as compared to "white box" statistical models, (b) have limited abilities to identify possible causal relationships, (c) model development is empirical; providing low decision insight, and many methodological issues remain to be solved, (d) models prone to overfitting, (e) require lengthy development and time to optimize, (f) they are more difficult to use in the field because of computational requirements, and (g) there is conflicting evidence as to whether or not they are better than traditional regression statistical models for either data classification, or for predicting outcome [21,45,46]
Summary
Artificial neural networks (ANNs) are regression devices containing layers of computing nodes (crudely analogous to the mammalian biological neurons) with remarkable information processing characteristics. The important milestone was the development of the first trainable network perceptron by Rosenblatt, 1959 [4] and Widrow & Hoff, 1960 [5], initially as a linear model having two layers of neurons or nodes (an input and an output layer) and a single layer of interconnections with variables (weights) that were adjustable during training. Only if the testing set has been used to set the network's weights or evaluate its structure, will it reflect the network's performance on future data; this practice of splitting the data into a training set and a test set is referred to as "cross validation" [18] Another method for estimating the error rate of a prediction rule is "data splitting" [19]. Going beyond triples may require excessive computational capabilities [24]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
