Study on the Efficiency of a Multi-layer Perceptron Neural Network Based on the Number of Hidden Layers and Nodes for Diagnosing Coronary- Artery Disease

Abstract

Background: Through the diagnostic decision support systems, potential patients or those who are on the threshold succumbing to a disease can be diagnosed early; thus, the prevention of unnecessary angiography for people not suffering from the coronaryartery disease as well as its dangers and costs can be avoided. The present study aimed at the efficiency evaluation of a multilayer perceptron neural network based on the number of hidden layers and nodes to diagnose coronary heart disease. Methods: A fundamental analysis was conducted on the provided data related to 13,228 patients who had undergone coronary angiography and the database (nine risk factors including age, gender, BMI, body fat, family history, smoking, blood cholesterol, diabetes, and high blood pressure) was investigated in this research using SPSS statistics (17.0) and R (2.13.2) software. In the next stage, through utilizing MATLAB (R2014a), 1332 different MLP neural networks were created. Results: Based on the largest area under the ROC curve, the best model of MLP neural network was selected involving two hidden layers; the first layer had 34 and the second one had 18 hidden nodes. This model had the highest efficiency of 82% in the diagnosis of coronary artery disease. Conclusions:The obtained results demonstrated that the MLP makes an acceptable approach to the diagnosis of coronary artery disease in patients without the need for performing angiography. The development of this model will result in creating an algorithm for decision support systems to diagnose coronary artery disease, as well.

Comparative Study of Classification Algorithms for Various DNA Microarray Data

“…Thus, the range of the parameter values and the optimal k value were different for each model. The MLP classification model generated two hidden layers since it has been verified effective in other studies for disease diagnosis [ 46 ]. Each hidden layer contained 10 neurons and used the ReLU activation function.…”

Deep Learning, a Not so Magical Problem Solver: A Case Study with Predicting the Complexity of Breast Cancer Cases

“…In MLP, data is handled one way and passes all layers once, as opposed to the way data is handled in recurrent neural networks. Back-propagation allows estimating what constitutes erroneous values in the hidden layer (which has no reference in the initial data to compare itself to) from the final classification labels [4].…”